Apoptosis of the midline glia during Drosophila embryogenesis: a correlation with axon contact

Cell death is a major feature during development of the central nervous system (CNS) in both vertebrates and invertebrates (Oppenheim, 1991; Bate et al., 1981; Truman et al., 1992). Although death has been noted in both neuronal and glial cell populations, neuronal cell death has received closer attention (Oppenheim, 1991). There is growing evidence, however, that glial cell number is also regulated by differential cell proliferation and cell death, leading to speculation that glial cell number may be regulated by mechanisms similar to those described for neuronal cell number (Raff et al., 1993).

In some neuronal populations, cell death results from the matching of cells to their efferent targets and is believed to operate by a trophic mechanism (Hamburger and Oppenheim, 1982). The deletion of targets result in reductions in the survival of the innervating population of neurons (Oppenheim, 1985; Hamburger and Oppenheim, 1982). Considerable evidence indicates that trophic support may also originate from afferent inputs, glia and the extracellular matrix (Okado and Oppenheim, 1984; Sohal et al.,, 1986; Furber et al., 1987; Walicke, 1989; Johnson and Deckwerth, 1993), and is modulated by both neurotrophins and afferent activity (Cunningham et al., 1979; Sendtner et al., 1990; Yan et al., 1992; Neff et al., 1993).

Death of ensheathing-type glia such as oligodendrocytes and Schwann cells has also received attention in the developing nervous systems of vertebrates (Knapp et al., 1986; Barres et al., 1992a; Raff et al., 1993; Louis et al., 1993; Doyle and Colman, 1993). Survival of newly formed oligodendrocytes in culture can be promoted by growth factors produced by their neighbour astrocytes or by growth factors present in the optic nerve (Louis et al., 1993; Barres et al., 1992b). In addition, there is evidence that glial-axonal contact plays a role in the survival of oligodendroglia. Oligodendroglia degenerate in neonatally transected rat optic nerves (David et al., 1984; Raff et al., 1993). Raff et al. (1993) proposed a model describing an early dependence of oligodendroglia on growth factors for differentiation followed by a dependence on axon contact for survival. The limited availability of axon-derived survival signals would serve to match the number of oligodendroglia to the length of axons requiring myelination.

Glial cell death in insects contributes to the remodeling of the central nervous system (CNS) during metamorphosis. For example, glial cell death occurs in the outer layer of the perineurium and the glial cover of the larval and pupal neuropil in the thoracic ganglia of Manduca sexta (Cantera, 1993). Natural glial cell death has been observed in metamorphic adults of the insect Manduca sexta where afferent input from antennal axons is required for the survival of glial cells in the glomeruli (Tolbert and Oland, 1989; Oland and Tolbert, 1987).

The midline glia (MG) of the midline of the Drosophila nerve cord participate in the morphogenesis of the commissural tracts during early embryogenesis (Jacobs and Goodman, 1989; Klämbt et al., 1991). Early MG survival and commissure morphogenesis requires the function of the rhomboid and Star genes (Sonnenfeld and Jacobs, 1994). Here we report that subsequent to their function in commissure morphogenesis, 50% of the Drosophila embryonic MG in normal embryos undergo apoptosis. The extent and time course of MG apoptosis is altered in mutant and transgenic embryos, which modify the degree of commissural axon contact with the MG. One possible explanation is that CNS axons may play a role in regulating the number of MG that survive to hatching. MG number may therefore be regulated during later embryogene-sis by trophic mechanisms.

AA142, a p-element insertion at 66D, labels the midline glia (MG) (Klämbt et al., 1991). The promoter fusion construct, slit-lacZ 1.0, is described in Wharton and Crews (1993) and was kindly provided by S. Crews. The rhomboidP38 (rhoP38) allele was generated from excision of a p-element insertion into 62A (Freeman et al., 1992). The slitIG107 (sliIG107) allele (Nüsslein-Volhard et al., 1984; Rothberg et al., 1988) and two deficiencies that uncover reaper (Df(3L)WR4 and Df(3L)WR10;White et al., 1994) were obtained from the Indiana stock centre. The commissureless1 (comm1) allele was kindly provided by C. S. Goodman (Seeger et al., 1993). Flies carrying a HS-rho p-element on the X chromosome (HS-rho-1B) were generously provided by E. Bier (Sturtevant et al., 1993).

Embryos were staged according to Campos-Ortega and Hartenstein (1985), incorporating three subdivisions of stage 12 (Klämbt et al., 1991).

Embryos containing a P[lacZ] element were incubated with a mono-clonal antibody against β-galactosidase (Sigma) which was detected using HRP cytochemistry according to Patel et al. (1987) and Grenningloh et al. (1991). A mouse polyclonal antibody to Drosophila peroxidasin (α-X) was used to identify phagocytic haemocytes (kindly provided by L. Fessler; Abrams et al., 1993; Tepass et al., 1994) which were visualized by HRP cytochemistry. The monoclonal antibody BP102 was used to visualize commissural and longitudinal CNS axon tracts (kindly provided by C. S. Goodman).

β-galactosidase histochemistry using X-gal (5-bromo-4-chloro-3-indolyl-B-C-galactopyranoside) and bluo-gal (halogenated indolyl-B-D-galactoside) was performed according to Jacobs et al. (1989) except that embryos were reacted for 12-16 hours at 18°C. For combined immunolabeling with X-gal and anti-peroxidasin, embryos were fixed in 4% formaldehyde.

Embryos containing AA142 and sli-lacZ 1.0 and reacted with X-gal were embedded and processed for light level microscopy according to Jacobs et al. (1989). Embryos reacted with bluo-gal were processed for electron microscopy as previously described (Jacobs, 1993).

Wild-type and HS-rhomboid (rho) embryos carrying the AA142 enhancer trap were collected on apple juice agar plates for 3 hours at 25°C. These embryos were then aged for 2 hours at 25°C and the plates were transferred to a 37°C water bath for 1 hour. The embryos were then allowed to develop for an additional 5 hours at 25°C or 12 hours at 18°C and were then processed for anti-β-galactosidase and BP102 staining. The ectopic expression of rho was verified by examining the wing phenotype (Sturtevant et al., 1993). To determine the effects of late ectopic rho expression, wild-type and HS-rho embryos expressing the AA142 enhancer trap were collected for three hours, aged for six to eight hours at 25°C and the plates were transferred to a 37°C water bath for one hour. These embryos were allowed to develop at 25°C for one hour and were then processed for anti-β-galactosidase and BP102 staining.

The midline glial lineage is derived from the mesectoderm of the Drosophila embryo. Three precursors divide once during stage 9 to produce three pairs of midline glia (MG), identified by position as anterior, middle and posterior (MGA, MGM and MGP, respectively) (Jacobs and Goodman, 1989; Klämbt et al., 1991). To determine glial function after commissure separation, we have used enhancer traps and a reporter construct to identify cells within the midline glial lineage during later embryogenesis. The enhancer trap AA142 is expressed in the MGA, MGM and weakly in the MGP in the embryonic CNS (Klämbt et al., 1991). The reporter fusion construct for the slit gene (sli-lacZ) is expressed in many MEC lineages prior to stage 13. Expression of sli-lacZ is restricted to the MGA and MGM during stage 15 and therefore is a useful marker of mature MG (Wharton and Crews, 1993).

We have observed variations in the position and number of MGA and MGM between ventral nerve cord (VNC) segments and embryonic stages, assessed with both AA142 and sli-lacZ 1.0 as shown in Fig. 1. Decreases in number and variations in position of the MGA and MGM were first observed during stage 13 and were present until the end of embryogenesis (Table 1; Fig. 1). Quantitative analyses revealed a 21% decrease in the number of MGA and MGM from stage 13 until stage 17 of embryogenesis. There are no changes in MG number from stage 17 (3.0±0.52 cells/segment, n=190) through first and second instar larva (L1 3.2±0.23 cells/segment, n=90). In 76.9% of segments more MGM than MGA survived while in 7.8% of segments more MGA than MGM survived. These results suggested that the MGM had a greater probability of survival than the MGA (Table 1). The MGM differs from the MGA because they contact axons of both the anterior and posterior commissure while MGA contacts anterior commissure axons only. Certain combinations of surviving MGA and MGM were not observed within VNC segments. For example, no segments were found to contain only one or no MG. Of the scored segments containing two MG, 3% of all segments scored contained one MGA and one MGM, 1.9% of all segments scored contained no MGA and two MGM, while no segments contained two MGA and no MGM (Table 1).

β-galactosidase expression of AA142 in the two MGP wasweak and transient and was excluded from cell counts in this analysis. The β-galactosidase staining in the MGP was diffuse during stage 13 (Fig. 1A) and was no longer detected after stage 14. The number of MGP was instead analysed using an enhancer trap (X55) that is expressed in the MGP, the ventral unpaired median neurons (VUMs) and the median neuroblast (MNB) (Klämbt et al., 1991). Two MGP arise from a midline glial precursor cell (Klämbt et al., 1991) and, by stage 14 of embryogenesis, there was an average of 1.6 MGP within VNCsegments (n=67) assessed with the X55 enhancer trap (Fig 2A). During stages 15 and 16 there was an average of 0.6 and 0.3MGP within VNC segments, respectively (n=100 and n=123;Fig. 2B). During stage 17 there was an average of 0.04 cells within VNC segments (n=178) while VNC segments (n=169) in early first instar larvae contained no MGP. Ultrastructural studies confirm that there is a reduction in the number of MGP after stage 14 (Jacobs, unpublished observation)

These quantitative analyses revealed that there was a combined 50% decrease in the number of MG from stage 13 to stage 17 of embryogenesis. We therefore investigated the possibility that the MG undergo natural apoptosis in the midline during embryogenesi

Initiation of apoptosis is blocked in embryos deficient for reaper (White et al., 1994). We examined X55 and sli-lacZ 1.0 expression in embryos homozygous for Df(3l)W4, which uncovers reaper, to determine whether the reduction in MG number could occur when apoptosis was blocked. In stage 16 embryos, 3-5 X55-expressing MGP (Fig. 2C), and 9-10 sli-lacZ 1.0 (Fig. 2D)-expressing cells were observed per segment. Small labeled profiles outside of the nervous system were not seen with either marker in Df(3l)W4 embryos. These results indicate that the decline in MG number during development is blocked in reaper-deficient embryos, and the deficiencies also result in a 60% increase in the number of cells expressing MG-specific markers.

We used electron microscopy to gain ultrastructural confirmation of midline glia cell death. Midline glia were identified with a substrate for β-galactosidase (bluo-gal) in sagittal sections from wild-type embryos carrying the AA142 enhancer trap. The bluo-gal substrate forms rod-shaped crystals around the nuclei of surviving and apoptotic midline glia (Fig. 3; Sonnenfeld and Jacobs, 1994). Surviving MG in wild-type embryos ensheath commissural axon tracts and have characteristic ultrastructural features such as extensive rough endoplasmic reticulum and electron-lucent cytoplasm (Fig. 3A). In wild-type embryos, midline glia labeled with the bluo-gal substrate were found in macrophages outside the dorsal surface of the VNC (Fig. 3B,A). These expulsed MG displayed features of apoptosis including condensed nuclei, reduced cytoplasmic volume and increased electron density (Figs 3B,4A) (Abrams et al., 1993). The nuclei of apoptotic glia in embryos mutant for the rhomboid (rho) gene are also found in macrophages dorsal to the VNC and possess similar apoptotic characteristics as the MG in wild-type embryos (Fig. 3C; Sonnenfeld and Jacobs, 1994).

To determine the time course of MG death during wild-type embryogenesis, we scored the number of whole-mount embryos containing apoptotic MGA and MGM from stage 12 until stage 17 using the AA142 enhancer trap as a cellular marker. Apoptotic MGA and MGM were first detected during stage 13 in 76.7% of scored embryos (n=34). During stage 14, 100% of embryos (n=57) contained apoptotic MG and by stage 17, 21.7% of embryos (n=23) contained apoptotic MG. AA142 embryos were double labeled with terminal transferase to identify when DNA fragmentation began (TUNEL, White et al., 1994). TUNEL-labeled apoptotic glia first appear at stage 13, in rounded cells already outside of the CNS. We conclude that DNA fragmentation was detectable at the same time that apoptotic cells can be identified by histological means. The temporal pattern of apoptosis within the MGP lineage was determined using the X55 enhancer trap. By stage 14, 83.3% of embryos (n=6) contained X55-labeled apoptotic cells and by stage 15 100% of embryos (n=9) contained apoptotic cells. By stage 17, 50% of scored embryos (n=16) contained X55-labeled apoptotic cells while no first instar larvae contained apoptotic cells (n=14). Although the X55 enhancer trap is also expressed in some midline neurons, other markers unique to those lineages (such as T13 and P223; Sonnenfeld and Jacobs, 1994) did not label apoptotic profiles.

In addition to the variation in MG position and number withinVNC segments of embryos, small AA142 and sli-lacZ-expressing cells were present in the haemolymph space dorsal to the VNC (Fig. 1A-D, arrowheads). These small displaced cells appeared to be contained within larger macrophage-like structures (Figs 1A-D, 4A). Identical results were observed using the X55 enhancer trap for MGP identification (data not shown). Two lines of evidence suggested that the small displaced AA142-, sli-lacZ- and X55-expressing cells were MG that had died and were expelled from the VNC: (1) the proximity of most profiles to the midline (Fig. 1G,H, arrowheads; Fig 4A) and (2) the fact that MG that undergo apoptosis in embryoswhich are mutant for the genes rho and Star (S) appear in similar positions (Sonnenfeld and Jacobs, 1994).

To confirm that the β-galactosidase-expressing cells in whole mount embryos were in macrophages, we used an antibody to peroxidasin, which is an extracellular matrix protein produced and secreted by macrophages (Abrams et al., 1993; Tepass et al., 1994). Expulsed MG were found in macrophages that were recognised by anti-peroxidasin (Fig. 4B-D). These expulsed MG were found in macrophages in various locations in the embryo including dorsal (Fig. 4B), lateral (Fig. 4D) and posterior to the VNC (Fig. 4C). All expulsed MG were located in macrophages outside the VNC as determined by histology (Fig. 4A; n=15) and whole-mount embryo analysis with the polyclonal antibody to peroxidasin (Fig. 4B-D; n=43).

Macrophages containing apoptotic MG were occasionally observed at considerable distances from the CNS, revealing mobility of the macrophages. Although we could not detect MG in the early stages of cell death by light microscopy, the fact that we rarely observed apoptotic MG within the VNC suggests that they were expelled before nuclear condensation was complete. Apoptotic MG are also rarely found within the VNC of embryos mutant for the rhomboid gene, suggesting that the expulsion of apoptotic MG before nuclear condensation may be a common occurrence (Sonnenfeld and Jacobs, 1994).

The identity of the surviving MG varied between embryonic VNC segments and stages suggesting that non-autonomous signal may be involved in MG survival. Axon contact has been proposed as a determining factor in the regulation of glial numbers in the rat optic nerve (Raff et al., 1993), and may explain some of the variation observed in the MG lineages. We therefore investigated the potential role of axon contact in regulating MG survival by analyzing MG fate in three contexts that altered MG-axon contact.

In commissureless (comm) mutant embryos, commissural axons rarely cross the midline of the VNC so that MG have minimal contact with axons (Seeger et al., 1993; Fig. 5A, B). The highest frequency of apoptotic MG (assessed with AA142) was observed in stage 13 comm mutant embryos (n=16) rather than stage 14 as observed in wild-type embryos. Apoptotic cellular profiles were no longer detected in stage 17 embryos (n=6). The temporal pattern of MG death in comm mutant embryos correlated with a 39% decrease in the number of MGA and MGM from stage 13 until stage 17 of embryogenesis, relative to a 21% decrease over the same period for wild type (Fig. 6). Surviving MG became laterally displaced towards the longitudinal axon tracts (compare Fig. 5A and 5B; Seeger et al., 1993).

The relationship between the MG and the CNS axons is altered in embryos mutant for the slit (sli) gene (Rothberg et al., 1988). In sli mutants, commissural and longitudinal axon tracts are collapsed at the midline (Fig. 5C,D; Rothberg et al., 1988). We have previously demonstrated that the MG in sli mutant embryos differentiate and ensheath axons at the ventral midline (Sonnenfeld and Jacobs, 1994). The temporal pattern of MG cell death in sli mutant embryos (assessed with AA142) was similar to that in wild-type embryos. In sli mutant embryos, there was a net 23% decrease in the number of MGA and MGM within VNC segments from stage 13 to 17 compared to a 21% decrease in wild-type embryos over the same period (Fig. 6).

It was possible to increase the number of MGA and MGM within VNC segments by ectopic expression of the rhomboid (rho) gene. Transgenic flies carrying a p-element containing the heat-shock (HS) promoter-70 fused to the rho gene (kindly provided by E. Bier;Sturtevant et al., 1993) were used to induce an increase in the number of MGA and MGM duringembryonic development of the CNS. MG were identified using the AA142 enhancer trap (Fig. 5E,F) and the sli-lacZ reporter (data not shown). Increases in the number of MGA and MGM were observed in stage 12 and stage 13 embryos that had been subjected to a 1 hour heat shock during stages 9 and 10 of embryogenesis (Figs 5E, 6). The increases in MG numbers were detected as early as the reporters are expressed. At stage 13 there was a 50% increase in the average number of MGA and MGM within VNC segments (Fig. 6). In embryos subjected to an identical heat shock but examined at later stages of development, the number of MGA and MGM declined towards the wild-type numbers of MG (Figs 5F, 6). The onset of observed MG death in HS-rho embryos was shifted earlier to stage 12 compared to stage 13 in wild-type embryos. It was therefore possible that the extra MG died by apoptosis and was reflected by a 40% decrease in the number of MGA and MGM from stage 13 until stage 17 of embryogenesis. The extra MGA and MGM did not perturb formation or separation of the anterior and posterior commissural axon tracts (data not shown).

The midline glia are responsible for the morphogenesis of the anterior and posterior commissures during embryonic development. A subset of the midline glia migrate between and separate the anterior and posterior commissures during stage 12, and subsequently ensheath axons of the commissural tracts (Klämbt et al, 1991). In the current study, we have shown that 50% of the MG died by apoptosis after the commissural axon tracts were separated. The use of enhancer traps allowed us to detect MG during late stages of apoptosis due to perdurance of alent in their developmental potential. Whether more MGM survive because they are necessary to stabilize commissure separation remains to be determined.

Apoptosis of the MG was blocked in embryos deficient for the gene reaper, previously established to be required for apoptosis (White et al, 1994). In the absence of apoptosis, there is an increase in MG number relative to the wild-type MG number. It is possible that these extra MG are derived from MECs that undergo apoptosis before glial-specific reporters can detect them. Alternatively, the absence of apoptosis may alter cell interactions during determination, thereby diverting extra cells to the MG fate. We did not observe apoptotic profiles in embryos carrying enhancer traps expressing β-galactosidase in the ventral unpaired median neurons (VUMs) and the MP1 and MP2 neurons (enhancer traps T13 and P223, respectively) (Sonnenfeld and Jacobs, 1994). These observations suggest that cell death within the mesectoderm may be limited to the three midline glial lineages.

Death of identified neurons in the embryonic grasshopper VNC can establish segmental specialisations and remove cells that have become obsolete (Bate et al., 1981). We did not find regional differences in MG survival between VNC segments. However, the correlation between the beginning of MG death and the separation of the anterior and posterior commissure suggests that fewer MG are required after this morphogenic event.

The MG are required for commissure separation and also ensheath these axons during embryogenesis (Klämbt et al., 1991; Jacobs and Goodman, 1989). Perhaps MG survival after commissure morphogenesis is modulated by contact with the commissural axons that they ensheath. We tested the hypothesis that axonal contact may be involved in MG survival by investigating three contexts where axon-glia contact had been altered. The decrease in contact between the MG and the commissural axons in commissureless (comm) mutant embryos was correlated with an 85% greater loss of cells from the MGA and MGM lineages. The decrease in axonal contact in comm mutant embryos was also correlated with an earlier peak in detectable apoptosis (stage 13) revealing that the MG may begin to respond to axon contact by late stage 12.

Although CNS axon tracts are abnormal in embryos mutant for the gene sli, the MG maintain axonal contact throughout embryogenesis (Sonnenfeld and Jacobs, 1994). Our analysis showed that in sli mutant embryos there was a decrease in the number of MGA and MGM (23%) comparable to wild-type MG decreases (21%) by the end of embryogenesis. It is possible that the collapse of CNS axon tracts toward the midline in sli mutants did not prohibit the axon-glial contact that may be involved in MG survival. Ectopic expression of the rhomboid gene during stage 9 and 10 increased the number of mesectodermal cells expressing MG-specific genes. This increase was corrected between stages 12 and 17 by increased levels of apoptosis. Commissure morphogenesis was unaffected.

For all three situations, an alteration in the ratio of MGs to commissural axons is corrected by changes in the number of MGs that undergo apoptosis. The genes sli, rho and comm are all expressed in the mesectoderm however, so mechanisms autonomous to the MG could also account for these observations (Rothberg, et al., 1988; Bier et al., 1990; Goodman, personal communication). It is possible that comm is expressed by the MG themselves, and is necessary to maintain MG differentiation and survival, as previously established in rho and Star mutant embryos (Sonnenfeld and Jacobs, 1994). Alternatively, the absence of comm activity, or excess rho activity could interfere with the ability of the MG to ensheath axons, resulting in their expulsion from the CNS. Analysis of MG survival in embryos genetically mosaic for comm and HS-rho could resolve these alternates.

Recent studies show that a rise in the level of ecdysteriods at pupariation controls the cessation of midline glial proliferation during the third larval instar, as well as triggering the eventual programmed cell death of the MG after 50% of metamorphosis (Awad and Truman, personal communication). Similarly we have found that the maximum number of glia are required during embryonic commissure separation and after this morphogenetic event some of the MG die. In both situations, a subset of the glia are transient, acting to facilitate morphogenesis of the nervous system, after which they are disposed. This is analogous to the transient mouse subplate neurons, which provide the template for the lateral geniculate body projection to the visual cortex. After this projection is established, and cortical dominance columns are formed, the subplate neurons die (Allendoerfer and Shatz, 1994).

Studies on trophic modulators of neuronal survival have noted a non-linear relationship between the amount of afferent or efferent input and the number of surviving neurons (Oppenheim, 1991). Similar effects are emerging in studies of glial survival. Glial-axonal contact appears to modulate oligodendrocyte survival in neonatally transected rat optic nerves (David et al., 1984; Raff et al., 1993). However, not all oligodendroglia die after optic nerve transection (David et al., 1984). Similar results have been found by studying the effects of a lack of afferent sensory input on glial cell differentiation in the antennal lobes of Manduca sexta during metamorphosis (Oland and Tolbert, 1987). Interestingly, we have found that not all MG die in comm mutant embryos. We cannot exclude the possibility that minimal axon contact could influence MG survival in comm mutant embryos. Nevertheless, the relationship between numbers of MG and numbers of axons is not linear. There are likely other factors that also modulate MG number in later embryogenesis.

A model by Raff et al. (1993) proposes that as oligodendroglia precursors differentiate into oligodendrocytes they lose sensitivity to certain growth factors. They then become dependent on axon-derived signals for survival. During a critical period, 50% of the oligodendroglia contact a myelinfree region of axon, and the remainder die. The competition for axon-derived signals may match the number of oligodendrocytes to the length of axons requiring myelination. We propose that a similar mechanism may function in Drosophila to regulate MG number.

Genes of the spitz group, including rho, are hypothesized to function in a signal transduction pathway (Rutledge et al., 1992; Sturtevant et al., 1993) and are required for MG survival after determination (Sonnenfeld and Jacobs, 1994). The increase in MG numbers generated by ectopic rho expression supports a role for the rho gene in MG determination. The midline cells are not responsive to ectopic rho expression after stage 13 of embryonic development, or once MG migration is complete. It is possible that once the MG complete their migration they become dependent on axon contact for survival. The correlation between decreased axon contact and decreased MG survival in comm mutant embryos supports this model.

These studies reveal that interactions between neurons and the MG may regulate glial cell number during embryonic insect development in a manner similar to that described in some vertebrate systems. The embryonic MG may serve as a useful in vivo model system to identify and characterise the role of trophic factors involved in glia survival.

We thank Lise Fessler, Ethan Bier and Corey Goodman for essential reagents. We are grateful to Ana Campos and Corrinne Lobe for constructive criticism of this manuscript. We thank Christoper Stemerdink for technical assistance. This work was funded by grants from the MRC and the Multiple Sclerosis Society of Canada.