An axon scaffold induced by retinal axons directs glia to destinations in the Drosophila optic lobe

Migration is a striking example of complex cellular behavior that is essential to development, wound healing and cellular immunity. In the nervous system, cell migration is a common step in the final positioning of neuronal and glial cell bodies prior to the assembly of neural circuitry. In the developing mammalian brain, neuronal precursors migrate along radial and tangential pathways to arrive at stereotyped destinations in the developing telencephalon (reviewed by Marin and Rubenstein, 2001). Schwann cell precursors migrate along axon tracts into the peripheral nervous system prior to ensheathing axon fibers(Carpenter and Hollyday, 1992). Given the central role of migration in neural development, it is not surprising that migration defects are common in human diseases of cortical organization (Ross and Walsh,2001). Nonetheless, the mechanisms by which migrating cells choose specific pathways and locate appropriate destinations are not well understood.

Drosophila melanogaster has been a favorable model system in which to study cell migration. Genetic analysis has been applied to the mechanisms controlling movement of follicle cells around the developing oocyte (reviewed by Montell, 1999) and gonadal precursors to the site of the developing gonad(Deshpande et al., 2001; Stein et al., 2002). In the nervous system, there are many instances of neuronal and glial migration from sites of cell birth to the locales of developmental and mature function, both in the CNS (Klambt et al.,1991; Klambt,1993) and PNS (Giangrande,1994; Choi and Benzer,1994; Sepp et al.,2000).

It is evident that the migration of neural cells to proper destinations can be a complex process, even in the relatively simple nervous system of the fruit fly. The adult Drosophila visual system contains glia of many distinct types and properties that are localized to specific sites in the retina and optic lobe (Tix et al.,1997; Saint Marie and Carlson,1983; Trujillo-Cenoz,1965). Some glia subtypes arise from progenitors located at sites well removed from their mature positions. Retinal basal glia arise from progenitors located in the optic stalk, the epithelial tube that connects the developing eye and brain. Basal glia precursors migrate to the basal surface of the retina (Choi and Benzer,1994; Rangarajan et al.,1999). In the optic lobe, lamina epithelial and marginal glia migrate to their specific layers from progenitor zones located at the prospective dorsal and ventral margins of the ganglion(Perez and Steller, 1996; Huang and Kunes, 1998). This migration depends on local signaling from the photoreceptor axons, as it largely fails to occur in mutants that do not produce photoreceptor neurons(Perez and Steller, 1996) or are deficient in the activity of the COP9 signalosome(Suh et al., 2002). Conversely, lamina glia are required for the guidance of photoreceptor axons,and provide an essential stop signal for the R1-R6 subset of photoreceptor axons to terminate their outgrowth in the lamina(Poeck et al., 2001). These complex processes are controlled with precision in the three-dimensional milieu of the developing brain by mechanisms that remain unclear.

An important aspect of glial cell migration is the targeting of particular subtypes to specific layers of the developing ganglia. Given their common origin in dorsal and ventral margin progenitor zones, selective mechanisms would be needed to direct glia to different destinations. We show that specific highways for migration are provided by axons that originate in the vicinity of glial progenitors. Photoreceptor axons induce the outgrowth of these `scaffold axons' upon their arrival in the brain, and targeted migration follows. These observations describe a cellular mechanism for the control of optic lobe glial migration by photoreceptor axons, an important element in the coordinated assembly of the precise neural architecture of the optic lobe.

All strains were grown on standard cornmeal medium(Ashburner, 1989). The following strains and transgenic animals were used in this study: wg-lacZ; so¹; hh¹; dpp-GAL4, UAS-CD8::GFP; omb-GAL4, UAS-CD8::GFP; ds¹; ds^33k; ds-lacZ; ey^D; UAS-Ras1^N17.

Third larval instar-staged animals were examined immunocytochemically essentially as described by Kunes et al.(Kunes et al., 1993). Primary antibodies were used at the following dilutions: mouse anti-Repo 1:10, goat FITC anti-HRP (Cappel) 1:200, rabbit anti-β-gal (Promega) 1:500, mouse anti-glial cells missing 1:100, monoclonal rabbit anti-human caspase 3 (BD PharMingen) 1:100 and rat anti-Elav 1:25. Secondary antibodies were used at the following dilutions: Cy3 or Cy5goat anti-mouse (Jackson Immunochemical,Inc.) 1:100, Cy3 or Cy5-goat anti-rabbit (Jackson) 1:500, HRP-conjugated goat anti-mouse IgG (Jackson) 1:100. Specimens were viewed on a Zeiss LSM510 META confocal microscope.

Mosaic analysis was carried out as described by Xu and Rubin(Xu and Rubin, 1993). Larvae were subjected to heat shock at 37°C for 5-7 minutes 24-36 hours after hatching to induce expression of an hsFLP transgene. After growth at 20°C,larvae were dissected at late third instar stage and processed for immunohistochemical analysis. The following crosses and strains were used in the experiments described:

1. Identification of scaffold axons and temporal analysis of Wg⁺neurons: wg-lacZ/y⁺CyO.

2. Marker analysis of wg-domains: (A) y,hsFLP₁₂₂; wg-lacZ/y⁺CyO X y,w; UAS-CD8::GFP,tubα1>y⁺,CD2>GAL4/y⁺CyO; (B) Omb-GAL4, UAS-CD8::GFP/FM7 X wg-lacZ/y⁺CyO; (C) Omb-GAL4, UAS-CD8::GFP/FM7 X ds-lacZ/y⁺CyO; (D) Dpp-GAL4, UAS-CD8::GFP/TM6B, Tb X wg-lacZ/y⁺CyO;(E) Dpp-GAL4, UAS-CD8::GFP/TM6B, Tb X ds-lacZ/y⁺CyO; and (F) Repo-GAL4, UAS-CD8::GFP/TM3,Sb.

3. Dependence of scaffold axon induction on retinal innervation: (A) so¹::wg-lacZ/y⁺CyO X so¹/y⁺CyO; and (B) ey^D X wg-lacZ/y⁺CyO. Retina development was assessed in both so¹ and ey^D animals by examining ommatida formation in the eye disc. For each specimen, regional brain development was assessed with respect to normal, or a lack of, correlated eye development.

4. Analysis of scaffold axon outgrowth: (A) ds¹::wg-lacZ/y⁺CyO X ds¹/y⁺CyO; (B) ds^33k::wg-lacZ/y⁺CyO X ds^33k/y⁺CyO; (C) y,hsFLP₁₂₂; wg-lacZ/y⁺CyO X y,w, UAS-Ras^N17; tubα1>y⁺,CD2>GAL4; UAS-CD8::GFP.

5. Analysis of optic lobe apoptosis: (A) y,hsFLP₁₂₂;wg-lacZ/y⁺CyO X y,w, UAS-Ras^N17;tubα1>y⁺,CD2>GAL4; UAS-CD8::GFP; and (B) so¹::wg-lacZ/y⁺CyO X so¹/y⁺CyO; (C) wg-lacZ/y⁺CyO.

(`>' indicates the position of an FRT site in the respective construct.)

Quantification of apoptotic cells in sine oculis and wild-type animals relied on carefully matched confocal and developmental parameters. Optical slices (1.3 μm) were compared for age-matched specimens at a similar focal plane for all specimens analyzed. The number of apoptotic cell profiles in 50 μm² regions both proximal to the medulla neuropile and within the medulla cortex of sine oculis and wild-type animals were counted. Cortical neuron size (∼3 μm) and overall brain size did not differ between sine oculis and wild-type animals, so cell counts were not biased by any intrinsic differences in cell sizes between these two genetic backgrounds. Differences in the number of neuropile-proximal and cortical apoptotic cells in sine oculis animals relative to wild-type animals were evaluated for significance using Student's paired t-test.

The Drosophila optic ganglia are derived from a pair of neuroectodermal proliferative centers known as the inner and outer anlage,which are located on the ventrolateral surface of each larval brain hemisphere(Green et al., 1993; Hofbauer and Campos-Ortega,1990; Nassif et al.,2003) (reviewed by Meinertzhagen and Hanson,1993). We have previously described the onset of Wingless (Wg)expression in a few cells located at the presumptive dorsal and ventral margins of the disk-shaped outer anlagen in young larvae(Fig. 2A)(Kaphingst and Kunes, 1994). These two `Wg domains' are positioned as adjacent wedges within the disk-shaped outer anlage (Figs 1, 2). In mid-metamorphosis, the disk opens at a fissure located between the two Wg domains (yellow bar in Fig. 1A); the disk unfurls linearly, such that the Wg domains move to the future dorsal and ventral poles of the ganglion. Interestingly, the Wg domains are roughly coincident with sites revealed by clonal analyses to be the origins of glia that migrate into the lamina neuropile (Perez and Steller,1996). We sought to investigate this relationship in greater detail.

We used a marker of wingless expression, the lacZreporter construct 17en40 (or wg-lacZ)(Kassis et al., 1992) to examine wg expression with respect to glial cell development and movement. In cells expressing wg-lacZ, cytoplasmicβ-galactosidase fills cellular processes and permits the visualization of their detailed morphology. We observed that the paths of glial cell migration coincided with a stereotyped pattern of cytoplasmic extensions emanating from wg-lacZ expressing cells (Fig. 1C,D,F, arrowheads). The extensions followed stereotyped routes and terminated at specific destinations in the lamina, medulla and lobula. Glia formed chains along these extensions and accumulated in neuropile destinations beyond the extension's termini(Fig. 1D). Marker analysis indicated that the glia undergo differentiation as they progress along these paths. Lamina glia express the early glial differentiation factor glial cells missing (gcm) (Hosoya et al., 1995; Akiyama et al.,1996; Egger et al.,2002) at their points of entry onto the extensions(Fig. 1E, white cells between arrowheads). Further along the path toward the neuropile, glia commence expression of the homeodomain protein Repo(Xiong et al., 1994) a marker downstream of Gcm expression. Thus, in the case of lamina marginal glia shown in Fig. 1E (Gcm expression) and Fig. 1D (Repo expression),differentiation can be assessed relative to their progression along the migratory pathway.

The wg-lacZ positive extensions are evidently axons. Their cell membranes were labeled by anti-horseradish peroxidase antibody, a neuronal marker (Fig. 1C)(Jan and Jan, 1982). Their cell bodies were labeled by Elav, also a neuronal marker(Fig. 1F)(White et al., 1983). The axons extending toward the same neuropile destination were bundled together in a fascicle, as indicated by labeling of individual axons in mosaic animals(described below; see Fig. 4). Four wg-lacZ labeled fascicles were resolved emerging from each of the dorsal and ventral Wg domains (Fig. 1A,B). One fascicle from each domain extended to the border of the lamina field, terminating at a position adjacent to layers of glia known as the lamina epithelial (Ep) and marginal (Ma) glia [established nomenclature(http://www.flybrain.org,Accession Number PP00003)] (Tix et al.,1997). These layers of glia lie, respectively, above and below the layer the axon termini of the R1-R6 photoreceptors. Glia could be observed migrating in a chain along the fascicles(Fig. 1D, between arrowheads). Owing to the absence of specific markers, we could not distinguish epithelial and marginal glia prior to their separation into distinct layers. It seems likely, however, that both glial types migrate on the same pathway. One fascicle from each Wg domain extended to the cortex, neuropile boundary of the medulla, and was associated with the chain-like migration of medulla neuropile(MNG) glia. One fascicle from each domain extended to the boundary between the medulla and lobula, and corresponded to a pathway for migration of inner chiasm (Xi) glia, which demarcate the border between medulla and lobula neuropiles (Tix et al., 1997). The final pair of fascicles extended into the lobula neuropile(Fig. 1F) and formed a pathway associated with lobula neuropile glia (LoG). We will refer to these four sets of putative migratory guides as `scaffold axons'. Notably, the migration of these four glia types depends on retinal innervation of the optic lobe, a requirement that is explored below.

Earlier stage specimens were examined in order to determine the temporal relationship between glial migration and the outgrowth of scaffold axons. During the first two larval stages, most neuroectodermal cells of the outer anlagen express the cell adhesion protein Fasciclin 2 (Fas2), with the exception of those in the two Wingless domains; Wingless-expressing cells are Fas2 negative (Kaphingst and Kunes,1994; Dumstrei et al.,2003). As development proceeds to the third instar stage, the neuroectodermal populations mostly convert to blast cells that produce the neurons and glia of the lamina and medulla. The differentiation of wg-lacZ positive scaffold neurons and the extension of their axons follow this general temporal scheme (Fig. 2). As seen in Fig. 2A,B, a small population of glia precedes the arrival of the first photoreceptor axons in the target field of the retinal axon. Perez and Steller(Perez and Steller, 1996) also described a small population of centrally located glia that preceded the arrival of photoreceptor axons in the lamina. At these early time points, when the Wg domains consist of fewer than a hundred cells; scaffold axon fascicles were not detected. Photoreceptor axons subsequently arrive in the temporal order that follows the posterior to anterior pattern of eye development. The elaboration of wg-lacZ positive scaffold axons was first detected as the first photoreceptor axons arrived in the target field, when only the first one or two columns of ommatidia had initiated differentiation in the eye disc(Fig. 2C). As retinal innervation continued, the number of glia in the neuropile regions increased steadily (Fig. 2D,E). Thus,Wingless-positive cells, though present long before the arrival of photoreceptor axons in the brain, appear to first extend axons when retinal axons begin to arrive in the brain.

The wg-lacZ labeled-axons also were examined in pupal stage animals, where mature axon projections into the optic lobe neuropiles could be resolved (Fig. 3). wg-lacZ positive axons could still be detected extending from cell bodies located at dorsal and ventral cortical positions. The axons projected to respectively dorsal and ventral targets in the medulla and lobula neuropiles (Fig. 3A,B). It is evident that the wg-lacZ-positive neurons include intrinsic neurons of the proximal medulla (Pm neurons), which extend arbors tangentially within small regions of the proximal medulla(Fischbach and Dittrich,1989). Projections into a specific tangential layer of the lobula were also observed. Projections into the lamina neuropile were not observed;at the third instar stage the extensions terminate at the border of the developing lamina neuropile. These axons thus may not have become part of lamina circuitry, or alternatively have ceased wg-lacZ expression at the pupal stages examined.

We sought to determine the location of progenitors that give rise to the distinct types of migratory glia and the neurons that formed their migratory pathways. The Wingless expressing cells of the dorsal and ventral domains are located in areas of complex gene expression controlled by Wingless (Wg)signaling activity (Fig. 4A)(Kaphingst and Kunes, 1994; Song et al., 2000). Adjacent to the Wg domains are non-overlapping cell populations that express the TGF-β family member Decapentaplegic (Dpp; Fig. 4C,E). Both the Wg- and Dpp-positive cell populations express the transcription factor Optomotor Blind(Omb; Fig. 4D)(Poeck et al., 1993; Huang and Kunes, 1998; Song et al., 2000). Dachsous(Ds), a Cadherin family member (Clark et al., 1995), is expressed a graded fashion with respect to the Wg domains (Fig. 4D,E)(Song et al., 2000). These three genes, though expressed in different patterns, are under the control of Wg activity (Song et al.,2000).

By performing clonal analysis with tissue labeled to provide positional landmarks, it was possible to localize glial progenitors and scaffold neurons to distinct sites of origin (Fig. 4B; Table 1). The FLP/FRT system (Golic and Lindquist,1989; Xu and Rubin,1993) was used to generate somatic clones that were positively marked by membrane-bound GFP (UAS-CD8::GFP)(Lee and Luo, 1999). Rare recombination events were induced such that most specimens harbored only one or a few labeled cell clones in the developing optic ganglia, which were examined in late third instar larvae. As reported by Perez and Steller(Perez and Steller, 1996),clones including lamina marginal and epithelial glia were found to label progenitors located at the dorsal and ventral margins of the outer anlagen(Fig. 4F,I). Clones that included lamina epithelial glia, lamina marginal glia, medulla neuropile glia,inner chiasm glia or lobula neuropile glia were all found within the domain of cells that express Wg, Omb and Ds (Domain I; Fig. 4A; Table 1). A majority of these clones contained both labeled scaffold neurons and glia (16/26); clones with only neurons or glia were less frequent. In the majority of specimens in which glia were labeled, the clone extended into multiple domains and included Domain I. With rare exception, these larger clones also contained neurons. Thus, at the time that somatic recombination was induced (the mid-second instar, 75 hours AEL (after egg laying)), most progenitor cells retained the potential to produce both neurons and glia. When the specimens were analyzed with respect to the glial types that were labeled, an interesting pattern emerged with regard to the position of labeled progenitor cells in the Domain I region. Labeled progenitors for each of glial cell type appeared in distinct domains on the proximal, distal axis (Fig. 4B). For example, the lamina epithelial and marginal glial progenitors were found in a more lateral position(Fig. 4F) than medulla neuropile glial progenitors (Fig. 4G). Inner chiasm glial progenitors were observed in an even more proximal location (Fig. 4H). Hence, the progenitor domains for distinct glial types appear to be organized into a proximal distal stack within Domain I, as illustrated in Fig. 4B.

Labeled scaffold neurons were most often included in the glial clones, and were found in close proximity to glial progenitors that used the axons as migratory guides (Fig. 4I,J). In a minority of cases, small clones were recovered which included only scaffold neurons (Fig. 4K). These clones were contained within Domain I (Wg, Omb, Ds expression) in all but two of nineteen cases. Our data do not resolve when the glial and neuronal lineages diverge. However, they do permit the conclusion that glia and the neurons that appear to establish their migratory pathways are generated in close proximity and with a lineage relationship.

The migration of glia from the prospective dorsal and ventral margins of the developing optic lobe depends on the arrival of photoreceptor axons in the target field (Perez and Steller,1996; Huang and Kunes,1998). When photoreceptor axons are absent, as occurs in mutants that eliminate ommatidial development (sine oculis, eyes absent and eyeless) most glia remain stalled in their progenitor domains(Perez and Steller, 1996). When photoreceptor axons innervate only part of the lamina field, glia migrate to the region that receives retinal innervation. It has thus been supposed that photoreceptor axons attract glia into the lamina target field. We thus thought it might be informative to examine the glial migratory scaffold under conditions where glial migration did not occur.

To this end, the axon scaffold was examined in sine oculis¹ (so) and Eyeless^D(ey) animals. These mutants display variable penetrance, such that photoreceptor neurons can be completely absent, or develop in variably sized clusters in a particular region of the developing retina. A lack of retinal innervation has compound effects on optic lobe development. Lamina neurons fail to develop because of the absence of axon-borne signals(Selleck and Steller, 1991; Huang and Kunes, 1996). The medulla is greatly reduced in cell number by extensive apoptosis(Fischbach, 1983). Employing mosaic analysis, Fischbach and Technau(Fischbach and Technau, 1984)showed that so¹ acts in the eye to bring about these effects on the brain.

In our analysis, the scaffold axons were labeled by the expression of wg-lacZ (Fig. 5). When photoreceptor axons were absent, the scaffold axons were likewise missing(Fig. 5B). However, wg-lacZ positive cells seemed to be present in normal number in the Wg domains (compare Fig. 5A″ with 5B″), and expressed the neuronal HRP antigens (e.g. the ventral Wg domain in Fig. 5C). In prior work, a number of markers expressed in the vicinity of the Wg domains were expressed normally in the absence of retinal innervation (e.g. Dpp and Omb) (Huang and Kunes, 1996; Huang and Kunes, 1998). Therefore, we think it unlikely that retinal innervation is required for the differentiation of these optic lobe neurons. When the scaffold axons were absent in so¹ and ey^D animals, glia accumulated at the edges of the Wg domains near the point where they would have joined axon fascicles on paths toward neuropile destinations (arrowheads in Fig. 5B′,C′). Most so¹ and ey^D animals develop part of an eye, such that the corresponding optic lobe receives partial innervation. In these cases, photoreceptor axons project to appropriate retinotopic locations despite the absence of the usual array of neighboring axons (Kunes et al., 1993). In such specimens, scaffold axons were found only in parts of the brain that received retinal innervation and not in regions that completely lacked innervation (Fig. 5C). Thus,for example, in the specimen shown in Fig. 5C, photoreceptor axons innervate the dorsal half of the developing optic lobe where scaffold axons have extended to the medulla neuropile, and medulla neuropile glia have migrated to normal positions. The ventral half of the optic lobe by contrast lacked retinal innervation. Scaffold axons and medulla neuropile glia were correspondingly absent; glia stalled at their exit point from the Wg domain (arrowhead in Fig. 5C). A correlation between retinal innervation, scaffold axon extension and glial migration was found in each of 32 so¹ animals with partial eye development restricted to either dorsal or ventral regions. These observations thus argue strongly that scaffold axon outgrowth and glial migration depend on retinal innervation.

Retinal innervation might elicit scaffold axon outgrowth, thereby establishing a necessary pathway for glial migration. Conversely, glial migration, elicited by retinal innervation directly, might establish a necessary pathway for scaffold axon outgrowth. Two approaches were undertaken in an attempt to resolve this issue. In the first, scaffold axons were eliminated by neuronal expression of activated Ras1 (Ras1^N17) in order to determine whether glial migration would occur in their absence. In the second, mutant animals with misdirected scaffold axons were examined in order to determine whether migratory glia follow the aberrant axon projections. Both approaches indicated that scaffold axons are necessary as glial migratory guides.

In the first approach, we used the ectopic expression of a dominant negative Ras protein (Ras1^N17)(Lee et al., 1996), which was found to autonomously block the outgrowth of scaffold axons(Fig. 6D,E). The UAS-Ras1^N17 transgene was expressed in small somatic clones that were labeled by the co-expression of UAS-CD8::GFP. Occasional somatic clones of Ras1^N17-expressing cells within the Wg domain region resulted in failure of scaffold axons to extend toward glial destinations. By contrast, clones that labeled lamina epithelial, marginal or medulla neuropile glia did not affect their respective migration (data not shown). When the scaffold axons were absent, glia stalled prior to migrating into the neuropile regions, as they did in so¹ animals(arrowheads in Fig. 6D,E). That is, glia did not migrate when the scaffold axons were missing, even though retinal axons were present in the target field. These observations indicate that the scaffold axon fascicles are required as migratory guides for glial migration.

In the second approach, we examined mutants lacking the function of the Cadherin family member Dachsous (Ds) (Clark et al., 1995). As described above, Ds is expressed in a gradient that decreases with distance from the Wg domains(Song et al., 2000)(Fig. 4D,E). Ds is highly expressed by neurons in the Wg domains but not by migrating glia. Ds can mediate homophilic binding, and so one might suppose that the Ds gradient could provide positional information for axon outgrowth within the developing neuropiles.

Scaffold axons did indeed project aberrantly in the mutants ds¹ and ds^33k(Fig. 6B,C) and glia lined these aberrant pathways and accumulated in abnormal destinations. In the ds¹ homozygous animal shown in Fig. 6B, scaffold axons which should project toward the medulla neuropile and form a path for medulla neuropile glia (as shown in Fig. 6A) were found extending toward the lobula cortex. Glia formed a chain along the aberrant axons (Fig. 6B′) and accumulated abnormally in the lobula, while the number that reached the medulla was reduced. In the ds^33khomozygous animal shown in Fig. 6C, the dorsal scaffold axons bifurcate and appear to direct glia away from paths toward neuropile destinations. Mutations in combgap,a negative regulator of optic lobe ds expression(Song et al., 2000) resulted in similar axon misrouting and glia distribution abnormalities (data not shown). The correlation between patterns of scaffold axon misprojection and glia migratory paths is a further indication that scaffold axons serve as migratory guides.

Prior work revealed extensive apoptosis in the optic lobes of `eyeless'mutants of Drosophila (Fischbach and Technau, 1984; Fischbach 1983). In the mutant sine oculis (so), mosaic analysis has revealed that extensive cell death in the optic lobe is due to the lack of so function in the retina and not the brain. These and additional observations have led to the conclusion that the optic lobe phenotype of sine oculis is due to lack of retinal innervation. The`trophic' function of photoreceptor axons could be direct, via provision of a survival factor, or indirect, e.g. by eliciting the migration of glia that provide a survival factor. We explore these distinct hypotheses below.

To address the issue, wild-type and so¹ animals were examined for the onset of apoptosis by their expression of activated caspase,which can be monitored in Drosophila with an antibody against the activated human caspase 3 protein (Baker and Yu, 2001). In third instar stage wild-type animals, few cortical cells are labeled by the anti-Caspase labeling(Fig. 7A″). By contrast, so¹ third instar larval optic lobes displayed an increased number of caspase 3-positive cells throughout the medulla cortex(Fig. 7B). Putative apoptotic cells were concentrated in regions where glia were particularly few or absent. Caspase 3-positive cells were also particularly prevalent in cortical areas immediately adjacent to the neuropile (arrowheads in Fig. 7B′). As noted above, areas particularly deficient in glia also displayed an irregular neuropile structure. These observations were subjected to a quantitative analysis (see Materials and methods). Areas of 50 μm² in 1.3μm confocal optical sections of the medulla cortex were counted for the number of caspase 3-positive cells. In wild-type animals, an average of 0.6(±1, n=10) activated caspase-positive cells were found per 50μm² area. No regional differences in the density of apoptotic cells were observed in wild-type animals. so¹ specimens displayed an average of 5.5 (±2.9; n=37) apoptotic cells in 50μm² areas adjacent to medulla neuropile regions that lacked medulla neuropile glia. Elevated apoptosis, an average of 6.5 (±5.2 n=37) activated caspase-positive cells, was also observed in distal cortical cell populations whose axons would normally innervate glia-starved regions of neuropile. By contrast, cell populations adjacent to glia-rich regions of medulla in the same so¹ animals showed only 2.4(±2.5 n=37) apoptotic cells per 50 μm², while in the corresponding distal cortical cell populations that innervate these glia-rich regions, only 2.5 (±3 n=37) apoptotic cells per 50μm² were found. Thus, although so¹ animals displayed an overall increase in caspase-positive cells, the frequency was significantly greater in cell populations that innervated glia-starved regions of neuropile. The results of this analysis are statistically significant: for medulla neuropile proximal regions within so¹ animals a paired t-test analysis yielded a P-value of 1.3e^-06 (comparing glia-poor and glia-rich regions), while the same statistical analysis comparing cortical regions yielded a P value of 4.8e^-07. Our observations suggest that both local and long-range trophic cues are provided by glia to cortical neurons.

To determine whether the increase in apoptotic cells was due to the absence of retinal axons or the absence of glia, we used Ras1^N17 expression to block the migration of medulla neuropile glia along the scaffold axons, as described above. In the specimen shown in Fig. 7C, a small clone of cells expressing the UAS-Ras1^N17 and UAS-CD8::GFP transgenes is localized to the Wg domain region. Glia accumulated at the border of this region instead of migrating into locations adjacent to the medulla neuropile. Notably,photoreceptor innervation was not markedly affected in such animals. The cortical area adjacent to this glia-deficient region displayed an increased number of anti-caspase 3-positive cells (arrowheads in Fig. 7C″). Thus, the absence of glia, rather than photoreceptor axons, appears the more likely cause of extensive apoptosis and neural cell loss observed in eyeless strains of Drosophila.

Glia and neurons are known to use one another's extended cellular processes to guide their migration over long distances and through complex terrain. Well-known examples include the migration of cortical precursors along radial cell processes (Rakic, 1988)and the movement of Schwann cell precursors along axon tracts into the PNS(Carpenter and Hollyday, 1992). The directed migration of glia along axons has also been described in Drosophila (Choi and Benzer,1994; Giangrande,1994; Sepp et al.,2000). This mode of cell dispersal has thus been conserved across significant evolutionary distances.

We show in this report that Drosophila optic lobe glia use axon fascicles as migratory guides and that the extension of these axon fascicles is induced by the ingrowth of photoreceptor axons from the developing retina(Fig. 5). The migratory scaffold axons emerge from optic lobe regions that are in close proximity to sites where glial cells originate; both arise in the dorsal and ventral domains where cells express the morphogen Wingless (Figs 1, 4)(Kaphingst and Kunes, 1994). When the scaffold axons were eliminated by the autonomous expression of an activated Ras transgene, glia failed to migrate and stalled at the borders of their progenitor sites (Fig. 6). Extensive cortical cell apoptosis ensued(Fig. 7). When the scaffold axons projected aberrantly (in animals mutant for the cadherin Dachsous; Fig. 6), glia followed the aberrant routes to incorrect destinations. The longstanding observation that glial migration does not occur in eyeless mutant Drosophila(Perez and Steller, 1996; Huang and Kunes, 1998) might thus be explained by an indirect mechanism in which innervation controls the establishment of an axon scaffold necessary to direct glial migration.

The migratory scaffold axons were identified by their cytoplasmic expression of β-galactosidase from lacZ under the control of a wingless promoter (Kassis et al.,1992). The neurons are thus residents of the Wg domains, a point additionally supported by labeling small numbers of neurons that projected their axons toward glial destinations (Fig. 4). In total, four different wg-lacZ delineated pathways were identified. These appear to account for all the pathways taken by optic lobe glia that have been identified as migratory by clonal studies. Separate pathways were identified for medulla neuropile glia, lobula neuropile and inner chiasm glia. A single scaffold axon pathway was observed leading to the marginal and epithelial glial layers of the lamina, suggesting that both of these glial types follow the same pathway. Perhaps these glia become separated only on the interposition of photoreceptor R1-R6 growth cones as they arrive in the lamina. Whether the epithelial and marginal glia arise from distinct precursors that migrate on the same pathway is unclear. In all cases, glia were observed to form migratory `chains' along axonal extensions, resembling a similar organization of migratory glia on retinal axons en route from the optic stalk to the eye field (Choi and Benzer, 1994) and from midline progenitor sites to destinations in the PNS (Sepp et al., 2000). In pupal stage animals, the wg-lacZ labeled neurons were observed in dorsal and ventral cortical locations, sending projections into neuropile targets consistent with the patterns of glial migration. However, no axons from wg-lacZ positive neurons were observed extending into the lamina neuropile at this stage.

The optic lobe regions surrounding the Wg domains display complex patterns of gene expression, mainly because of the signaling activity of Wingless(Kaphingst and Kunes, 1994; Song et al., 2000). Our clonal analysis indicates that all five migratory glial cell types we examined arise from these domains (Fig. 4 and Table 1). Interestingly, the sites from which particular glia arise are stacked on the proximal distal axis(Fig. 4B) in a manner that correlates with target destinations in the developing ganglia. Thus, for example, somatic clones that label the medulla neuropile glia are located at a position that corresponds to the medial/distal position of MNG glia relative to the lamina and lobula glia. Furthermore, on the basis of their expression of wg-lacZ, as well as clonal analysis(Fig. 4), the neurons that extend scaffold axons arise in close proximity to the sites of glial origin. Indeed, as somatic clones induced in mid-second instar larval animals often labeled both scaffold neurons and migratory glia, the glia and neurons must share common progenitors. It is curious that all of these distinct cell types express wingless. We have no evidence that axonally transported Wg functions in optic lobe development, as it does in the development of the neuromuscular junction (Packard et al.,2002). Partial elimination of wg⁺ activity (by the use of a conditional wg^ts allele) did not result in a specific defect in glial migration in the optic lobe (R.D. and S.K.,unpublished) but other possible functions of Wg were not addressed by this analysis.

Early studies on sine oculis mutants(Fischbach, 1983; Fischbach and Technau, 1984)demonstrated that lack of retinal innervation results in excessive optic lobe cell death, especially in the medulla, which is reduced to less than half of its normal volume without innervation(Power, 1943). Our analysis suggests that these earlier observations may be accounted for by a lack of glial migration. Though apoptosis was generally elevated in the medulla of animals lacking retinal innervation, a higher level of cell death was found in regions particularly devoid of medulla neuropile glia(Fig. 7). This effect could be produced with retinal axons present when glial migration was blocked by elimination of the scaffold axons (via Ras1^N17 expression). Therefore, retinal innervation and its inductive effect on neuronal development in the lamina is not sufficient for survival. Glial migration is required. A role for glia in neuronal survival has also been documented in the lamina, where neuronal cell death follows the elimination of glia in animals carrying a visual system specific allele of repo(Xiong and Montell, 1995).

Our observations suggest a developmental mechanism for the control of glial cell migration that depends on the establishment of an axon scaffold for guidance of migrating glia (Fig. 8). In normal development, a small number of glia migrate into the target field of the photoreceptor axon prior to the arrival of the first photoreceptor axons (Fig. 2)(Perez and Steller, 1996). These glia, which migrate independently of retinal innervation, may serve a necessary early role in photoreceptor axon guidance. They may be targets for the first retinal axons to arrive in the optic lobe, and provide the first signals that differentiate the outgrowth termination points of the R1-R6 and R7/8 axons (Poeck et al.,2001). We propose that the first photoreceptor axons to arrive in the optic lobe elicit outgrowth of the scaffold axons from neurons at the dorsal and ventral margins. Subsequent migration of glia from the dorsal and ventral margin progenitors is then both permitted and directed along the specific pathways of the scaffold. After the erection of the migratory scaffold, glial migration may be independent of continued photoreceptor axon ingrowth. On this point, we note that glia migrate into the lamina in approximately normal number in hh¹ animals, in which ommatidial development ceases after 11-13 columns form at the posterior of the developing retina (Huang and Kunes,1998). Therefore, glial migration may not depend on continued arrival of new retinal axons in the lamina primordium. Nonetheless, this observation cannot rule out the alternative interpretation that retinal axons emit a continuous attractive signal for glial migration that functions in adjunct with the migratory axon scaffold.

How do photoreceptor axons control the outgrowth of axons from the Wg domains? This does not appear to be a consequence of an affect of retinal innervation on neuronal development in the Wg domains, which appear normal in size and organization in eyeless mutant Drosophila strains(Song et al., 2000). Hedgehog,which is brought into the brain by retinal axons, is not required for the expression of either Wg or Dpp at the dorsal and ventral margins of the optic lobe (Huang and Kunes, 1998). It seems that the induction of scaffold axon outgrowth by the photoreceptor axons may be direct, as the outgrowth occurs in the hedgehog mutant, hh¹, in which the first steps of lamina neuronal development fail to occur (Huang and Kunes, 1996). Thus, one might suppose that retinal axons emit a chemoattractant for scaffold axon outgrowth, either synthesized in the retina or acquired from environmental sources and redistributed by retinal axons(e.g. as for Netrin) (Hiramoto et al.,2000).

The system for glial migration guidance we describe permits diversified cell types to originate from a common site, and yet target specific locations in complex neuropiles. One could imagine that as more complex and diversified neuropiles evolved, relatively simple changes in the developmental pathway of glial progenitors and the projections of scaffold axons would deliver glial support to new structures. This system also has the feature of functioning as a developmental timing `checkpoint' that fine-tunes the general hormonal coordination of imaginal development. Thus, upon their initial arrival in the brain, retinal axons provide a fine level of local cellular control over the movement of glia, preparing the target field for the next steps of optic lobe development.

This work was supported by NIH-NEI grant EY10112 to S.K., and NIH-NEI grant EY07030 to R.D. We appreciate the receipt of strains and antibody reagents from Michael Hoffman (Wisconsin, Madison), Seymour Benzer (Caltech), the Developmental Studies Hybridoma Bank (Iowa) and the Bloomington Drosophila Stock Center.