Segment-specific prevention of pioneer neuron apoptosis by cell-autonomous, postmitotic Hox gene activity

During embryonic development, programmed cell death (PCD) plays a fundamental role in the precise regulation of cell numbers in a multitude of tissues (Jacobson et al.,1997). A large body of work during the last three decades has revealed a central role for a group of cysteine proteases known as caspases,whose activation in a cascade manner is crucial to the execution of apoptosis in all animals (Hengartner,2000). While this core apoptotic machinery has been remarkably well conserved throughout evolution, important differences between species have emerged regarding its control. Caspase activation in vertebrates and Drosophila is dependent upon activation of RHG-motif proteins, which bind to and thereby prevent the function of inhibitor of apoptosis proteins(IAPs) (Bergmann et al., 2003). The biological significance of this control is highlighted by the fact that in Drosophila, elimination of three of these five RHG-motif cell death activators results in complete absence of embryonic apoptosis(White et al., 1994). In contrast, these genes have not been identified in the C. elegansgenome (Baehrecke, 2002). Other aspects of PCD regulation are unique to vertebrates; for instance, while developmental apoptosis in invertebrate embryos (including C. elegansand Drosophila) typically occurs immediately after a cell is born(Lundell et al., 2003; Sulston and Horvitz, 1977; Sulston et al., 1983) this is often not the case in vertebrates. Target-derived neuronal survival in the developing vertebrate nervous system is a striking example; as a result of their failure to reach target-derived neurotrophic factors, between 50% and 80% of vertebrate embryonic motor neurons undergo apoptosis after extending their axons (Oppenheim, 1991). Thus, the regulators of the core apoptotic machinery and the mechanisms by which they control cell death have become increasingly sophisticated throughout evolution, and their study is pivotal to the understanding of how and when cells become determined to die.

Amongst such regulators are Hox genes, whose expression in specific domains along the antero-posterior (AP) axis is translated into axial morphological differences through their action on a number of biological processes(Mann and Morata, 2000; McGinnis and Krumlauf, 1992; Warren and Carroll, 1995),including apoptosis (Alonso,2002). Indeed, increased or reduced cell death has been observed in mouse, Drosophila and nematode Hox gene mutants(Economides et al., 2003; Lohmann et al., 2002; Sommer et al., 1998; Tiret et al., 1998). Furthermore, a direct, in vivo modulation of the apoptotic machinery by a Hox gene was recently reported; in Drosophila, the Hox gene Deformed (Dfd) promotes localized PCD in the embryonic head lobes by activating the transcription of the RHG-motif gene reaper(rpr) (Lohmann et al.,2002). Together, these results point to a link between Hox function and apoptosis, and suggest that the effect of Hox function on cell death is dependent on cellular context.

The nervous system is also regionally specified, and Hox genes have been shown to control its segmentation(Capecchi, 1997) and to specify neuronal identity along its anteroposterior (AP) axis(Dasen et al., 2003; Jungbluth et al., 1999; Liu et al., 2001). In the vertebrate spinal cord, axial differences in the size of embryonic motor neuron pools exist long before they become dependent upon target-derived signals (Brown, 1981). Similarly, in the developing Drosophila ventral nerve cord (VNC),segmental differences in the numbers of neurons and glia arise from an initially equivalent set of 30 neuroblasts in each hemisegment(Prokop and Technau, 1994; Schmid et al., 1999; Udolph et al., 1993). These differences are likely to result from differential Hox gene modulation of both proliferation and apoptosis along the AP axis(Prokop et al., 1998). Indeed,the recent finding that the Drosophila Hox gene abdominal A(abd-A) activates apoptosis of postembryonic neuroblasts, thereby regulating the final number of neurons in the adult fly, provided an elegant example of how Hox genes can act on neuronal precursors to generate axial diversity (Bello et al., 2003). However, differences in neuronal architecture along the AP axis could alternatively emerge from selective elimination of mature neurons in specific axial domains. In this model, an identical neuronal profile would initially be generated along the AP axis. Subsequently, patterning genes would act in their specific expression domains to differentially trigger or prevent apoptosis of mature cells.

In this study, we demonstrate the first example of apoptosis of differentiated neurons in an invertebrate embryo by showing that the Drosophila dMP2 and MP1 pioneer neurons undergo segment-specific apoptosis at a late developmental stage. These neurons are initially generated throughout the VNC (Bossing and Technau,1994; Doe, 1992; Schmid et al., 1999; Schmidt et al., 1997), extend their axons and perform a critical pioneering role in guiding follower axons(Hidalgo and Brand, 1997; Lin et al., 1995). We have found that at a later stage, dMP2 and MP1 neurons undergo apoptosis only in anterior segments. While neither the generation nor the initial function of these neurons depends on Hox function, their late, segment-specific survival is under homeotic control; the Hox gene Abdominal B (Abd-B)acts in posterior segments in a cell-autonomous and postmitotic fashion to prevent apoptosis of dMP2 and MP1 neurons by repressing two RHG-motif cell death activators, rpr and grim. Our findings provide clear evidence for a cell-autonomous and anti-apoptotic function of a Hox gene in vivo. Furthermore, they identify a novel mechanism linking Hox positional information to axial differences in neuronal architecture by the selective elimination of mature neurons.

The following fly stocks were used: Abd-Bm mutants: Abd-B^M1, Abd-B^M2, Abd-B^M5(Sanchez-Herrero et al.,1985); Abd-B mutant for both m and r isoforms: Abd-B^D18 (Hopmann et al., 1995); UAS-Antp, UAS-Ubx, UAS-abd-A(Hirth et al., 2001) (obtained from F. Hirth); UAS-Abd-Bm(Castelli-Gair et al., 1994)(obtained from J. Castelli-Gair); UAS-p35(Hay et al., 1994) (obtained from J. Nambu); XR38, H99, X14 and X25 deficiencies(Peterson et al., 2002)(obtained from K. White); elav^GAL4 (elavC155)(Lin and Goodman, 1994); HB9-GAL4 (Broiher and Skeath,2002) (obtained from H. Broiher); repo^GAL4(Sepp et al., 2001) and sim^GAL4 (Golembo et al., 1996) (both obtained from U. Gaul); UAS-myc-EGFP^F (Allan et al., 2003); UAS-nls-myc-EGFP(Callahan et al., 1998)(obtained from D. van Meyel). dMP2-GAL4 was initially identified as P0201 in the Flyview stock collection(http://pbio07.uni-muenster.de/FlyView/Home.html)and was described as Vap^GAL4(Allan et al., 2003). This GAL4 insertion maps to the third chromosome and its precise insertion site was determined by plasmid rescue (not shown). Since there is no difference in anterior cell survival between wild-type, heterozygous or homozygous dMP2-GAL4 larvae, we conclude that this Pinsertion does not affect dMP2 apoptosis. The rpr^GAL4 line(103634 NP: 520) was generated and mapped by the NP Consortium(http://flymap.lab.nig.ac.jp/getdb.html). We analyzed the expression of several previously generated rpr-lacZpromoter fusions (Brodsky et al.,2000; Jiang et al.,2000). In our hands, all transgenes (including apparently identical fragments) revealed different expression patterns in the stage 16 VNC, thus making it difficult to assess how well they mimicked the endogenous rpr expression (not shown). w¹¹¹⁸ was used as a wild-type strain. Mutations were maintained over standard balancers with GFP markers. Mutants were identified by the absence of GFPexpression or, where relevant, by other markers.

Immunolabeling was carried out as previously described(Allan et al., 2003). Antibodies used were: mouse α-myc mAb 9E10 (1:25), mouse α-Antp mAb 4C3 (1:50), concentrated mouse α-Abd-B 1A2E9 (1:5) (all from Developmental Studies Hybridoma Bank); rabbit α-Proctolin (1:2000)(Taylor et al., 2004), rabbitα-Odd (1:1000) (Spana et al.,1995), rabbit α-GFP (Molecular Probes, 1:1000), mouseα-Ubx mAb FP3.38 (1:20) (White and Wilcox, 1984), mouse α-Abd-A mAb DMabdA.1 subclone 6A8.12(1:400) (Kellerman et al.,1990). FITC-, Rhodamine-Red-X- and Cy5-conjugated secondary antibodies were obtained from Jackson Immunolabs and used at 1:200 (1:100 for the Cy5-conjugated antibody). Lipophilic DiI injections were done as previously described (Landgraf et al.,1997). Phalloidin-TX (Molecular Probes) was used at 1:1000. Where appropriate, images were false-colored for clarity, and red was converted to magenta for the benefit of color-blind readers.

We recently identified a set of six peptidergic neurons in the Drosophila larval VNC. Because these neurons appeared to be present only in the three posterior-most abdominal segments of the VNC (A6 to A8), we named them Vap (ventral abdominal posterior) neurons(Allan et al., 2003). This segmental pattern was surprising, given that the eight abdominal segments of the Drosophila VNC have traditionally been grouped into two genetically similar units: A1 to A7, and A8(Campos-Ortega and Hartenstein,1985). Since the segmental occurrence of the Vap neurons did not fit this conventional division, we sought to analyze their development in detail. For this purpose, we examined the expression of our previously described Vap-specific GAL4 line (Vap-GAL4)(Allan et al., 2003) throughout development. This revealed that while the Vap neurons were confined to the three posterior-most abdominal segments in larval stages(Fig. 1F), they were present throughout the VNC at embryonic stages 14-15, when Vap-GAL4expression was first detected (Fig. 1A). At these stages, both their cell body position (located medially and immediately underneath the posterior commissure), as well as their posterior axonal projections, were highly reminiscent of the extensively studied dMP2 neurons.

The dMP2 neurons are among the first neurons in the VNC to differentiate;they are generated at the end of embryonic stage 10 and extend their axons soon thereafter (stage 12) (Broadus et al., 1995; Doe,1992; Spana et al.,1995). Cell ablation studies have revealed that they play an important role as pioneer neurons that guide the axons of later-born neurons(Hidalgo and Brand, 1997; Lin et al., 1995). The dMP2 neuron and its sibling, the vMP2 neuron, arise from a single division of the MP2 neuroblast, which expresses the zinc-finger gene odd-skipped(odd). While Odd is transiently expressed in both dMP2 and vMP2 neurons, it becomes restricted to the dMP2 sibling from embryonic stage 12(Spana et al., 1995). Odd is additionally expressed in one other cell type in the VNC from this stage, the MP1 neuron, which can be distinguished from dMP2 by its smaller and elongated cell size, as well as its more dorsal location(Spana et al., 1995). One dMP2 and one MP1 neuron are generated in each hemisegment throughout the VNC(Schmid et al., 1999). Using Odd as a marker, we confirmed that the Vap neurons are indeed the dMP2 neurons(Fig. 1G,H). Thus, we will subsequently refer to the Vap-GAL4 line as dMP2-GAL4.

Because the dMP2 neurons extend axons posteriorly to pioneer the longitudinal tract (Thomas et al.,1984), they had been previously characterized as interneurons. However, analysis of their axonal projections using a membrane-targeted reporter (dMP2-GAL4/UAS-myc-EGFP^F), as well as lipophilic dye (DiI) backfills from the hindgut, allowed us to determine that posterior dMP2 `interneurons' actually exit the VNC and innervate the hindgut(Fig. 1J-L). Given that most,if not all, Drosophila motor neurons show signs of activated BMP signaling (Marques et al.,2002), this observation is in accord with our previous finding that the dMP2 neurons (or Vaps) express phosphorylated Mad (pMad)(Allan et al., 2003). The Drosophila hindgut is innervated by axons positive for the invertebrate myomodulator Proctolin(Anderson et al., 1988), and we found that the dMP2s indeed express this neuropeptide(Fig. 1I). Thus, the Drosophila dMP2 `interneurons' are, in fact, Proctolin-expressing peptidergic motor neurons that innervate the hindgut(Fig. 1M).

The absence of anterior (S3, T1 to T3 and A1 to A5) dMP2 neurons in larvae(but not in stage 16 embryos) could result from specific downregulation of dMP2-GAL4 expression anteriorly. Alternatively, it could reflect the selective death of anterior dMP2 neurons at a late embryonic stage. To resolve this issue, we used dMP2-GAL4 to examine these neurons at embryonic stage 17. Surprisingly, the dMP2 neurons displayed several of the morphological hallmarks of apoptosis; they appeared pyknotic (shrunken and densely stained) and their axons were fragmented(Fig. 1B-E). Pyknotic corpses seemed to be transported to the dorsal surface of the VNC(Fig. 1D, arrow), where dying cells have been reported to be engulfed by macrophages(Sears et al., 2003; Sonnenfeld and Jacobs, 1995). By late embryogenesis (air-filled trachea stage), all 18 anterior dMP2 neurons were lost (0% survival; Fig. 7B). In contrast, the six dMP2 neurons in segments A6 to A8(subsequently referred to as posterior dMP2s) survived and were maintained throughout the four-day larval period (Fig. 1F; not shown). We further examined the larval expression of Odd,as well as that of other previously described dMP2 markers – namely, 15J2-GAL4 (Hidalgo and Brand,1997) and AJ96-lacZ(Spana et al., 1995) –and found them to be absent from anterior segments (see below; not shown),further suggesting that anterior dMP2 neurons die at a late embryonic stage.

In Drosophila, programmed cell death is critically dependent upon a family of IAP inhibitors, the RHG-motif genes reaper(rpr), grim and head involution defective(hid) (Bergmann et al.,2003). Embryos homozygous for the chromosomal deletion Df(3L)H99 (H99) lack these three genes(Fig. 7A) and show an apparently complete absence of apoptosis(White et al., 1994). To confirm that anterior dMP2 were indeed undergoing apoptosis at embryonic stage 17, we examined late H99 embryos (air-filled trachea stage) and found that all anterior dMP2s survived (99%; Fig. 2A,B; Fig. 7B). To identify which of the three RHG-motif genes normally initiated anterior dMP2 death, we made use of additional deletions that remove one or two of these three genes (Peterson et al.,2002) (Fig. 7A). We did not observe any anterior dMP2 survival in embryos heterozygous for either gene alone (wt/X14, wt/X25 and wt/XR38, not shown) or in combination (wt/H99; Fig. 7B). However, the complete removal of rpr(XR38/H99) resulted in survival of 98% of anterior dMP2 neurons(Fig. 2C; Fig. 7B). Surviving anterior neurons expressed Odd (Fig. 2E)and, in some cases, Proctolin (Fig. 2G).

The XR38/H99 deletion combination removes rpr, but is also heterozygous for hid and grim(Peterson et al., 2002). Since cell death activators have been previously reported to act synergistically(Zhou et al., 1997), we wondered whether a partial absence of hid and/or grim might be contributing to the survival observed in XR38/H99 mutant embryos. Indeed, when we examined embryos lacking rpr but otherwise wild type for grim and hid (XR38/XR38), we observed that only 34% of anterior dMP2s survived (Fig. 7B). We then tested whether a complete loss of hid and/or grim function could affect dMP2 cell death and found that neither the loss of grim and/or hid alone (X14/X14 or X25/X5) resulted in anterior survival. However, complete loss of grim in a background heterozygous for rpr (X25/H99)resulted in survival of 66% of anterior dMP2s(Fig. 7B). In contrast, loss of hid in the same genetic background (X14/H99) had little effect on anterior survival (Fig. 7B). Thus, we conclude that anterior dMP2 death is primarily mediated by rpr with some contribution of grim.

To rule out the possibility that the persistence of anterior dMP2 neurons in the aforementioned mutant scenarios resulted from blocking cell death in a non-autonomous fashion, we used the dMP2-GAL4 line to express the baculoviral anti-apoptotic caspase inhibitor P35(Hay et al., 1994)postmitotically in all dMP2 neurons. This resulted in the survival of anterior dMP2 neurons in all VNC segments (Fig. 2D), with accompanying expression of Odd and, in some cases,Proctolin (Fig. 2F,H). This further supports that the death of anterior dMP2 neurons is apoptotic in nature.

Cell death of anterior dMP2s could be achieved by selective activation of rpr transcription anteriorly. Alternatively, rpr could be transcribed in both anterior and posterior dMP2s, with post-transcriptional mechanisms – or regulation of other genes downstream of rpr– accounting for the differential survival of anterior and posterior dMP2s. This issue was difficult to address since the embryonic expression of rpr is highly dynamic and there are no antibodies available with which to detect Rpr protein. However, a GAL4 P element transposon line inserted 125 base pairs upstream of the predicted rprtranscriptional start site was recently generated (subsequently referred to as rpr^GAL4; see Materials and methods). Since this insertion line allows for expression analysis with single-cell resolution, we used rpr^GAL4 as a probable readout of rpr expression. Its expression in the VNC is dynamic and commences at stage 15(Fig. 2I). Double labeling for rpr^GAL4 and Odd indicates that anterior, but not posterior, dMP2s express rpr^GAL4 at the time when they begin to die (Fig. 2J). This suggests that differential transcriptional activation of rpr in anterior dMP2s underlies their death. Anterior expression is variable and may reflect the fact that Odd expression is sometimes already lost by the time that rpr^GAL4 levels become high enough to be detected. In fact, we often observed cells with strong rpr^GAL4expression in positions where Odd-positive dMP2s would have been expected (not shown).

As expected from its insertion site, rpr^GAL4 is an allele of rpr; 32% of anterior dMP2s survive in rpr^GAL4/H99 embryos(Fig. 7B). The difference in dMP2 survival between the rpr^GAL4 allele and the XR38 deletion (32% versus 98% over H99) may indicate that rpr^GAL4 is not a null rpr allele. Alternatively,or additionally, XR38 may remove the recently identified RHG-motif gene sickle, located close to rpr but outside the H99 region (Christich et al.,2002; Srinivasula et al.,2002; Wing et al.,2002) (Fig. 7A). If that is the case, XR38/H99 would be heterozygous for skl,which might contribute to anterior dMP2 survival.

Based upon (1) the loss of Odd, 15J2-GAL4 and AJ96-lacZexpression in anterior dMP2 neurons at stage 17, (2) the apoptotic appearance of pyknotic dMP2 cell bodies and their fragmented axons as visualized by dMP2-GAL4/UAS-myc-EGFP^F at this stage, (3) the analysis of Df(3L)H99/XR38 and rpr^GAL4/H99 mutants, (4) the cell-autonomous rescue of anterior dMP2 obtained with dMP2-GAL4/UAS-p35 and (5) the expression of rpr^GAL4, we conclude that the Drosophila dMP2 neurons are generated in all VNC segments, that they differentiate and extend axons at stage 12, and that at stage 17 they activate rpr and undergo rpr- (and partly grim-) mediated programmed cell death in all VNC segments anterior to A6.

Why would only anterior dMP2s undergo apoptosis at stage 17? Since dMP2 neurons are motor neurons that exit the VNC, we initially speculated that they perhaps depend upon a target-derived signal. Anterior dMP2s would fail to exit the VNC on time to receive this signal and would therefore die. Although such survival signals have not been described in invertebrates, recent studies have revealed that both Drosophila motor neurons and peptidergic neurons depend upon target-derived BMP signals for both proper differentiation and maturation (Aberle et al.,2002; Allan et al.,2003; Marques et al.,2002; Marques et al.,2003). To test this idea, we carried out a number of mutant and tissue-ablation experiments (details available upon request). However, they all failed to lend support to this model.

We then hypothesized that the mechanism preventing posterior dMP2 death may be cell-autonomous. Recently, two studies have uncovered a pro-apoptotic function for Hox genes in Drosophila(Bello et al., 2003; Lohmann et al., 2002). While Hox genes often appear to be broadly expressed within their AP domain(Hirth et al., 1998), their precise expression patterns at the time when neurons and glia are generated have not been well characterized in Drosophila. For these reasons, we sought to examine in detail the expression of the Hox genes normally expressed in the Drosophila VNC, and to determine whether it may be of relevance to the segment specificity of dMP2 death.

When anterior dMP2s degenerate, the vast majority of embryonic neurons and glial cells have differentiated. We observed that at this stage, the Hox proteins Antennapedia (Antp), Ultrabithorax (Ubx), Abd-A and Abd-B were not ubiquitously expressed within their expression domains. For example, some postmitotic neurons appeared not to express specific Hox proteins(Fig. 3A-C, arrowheads). Furthermore, Hox expression was largely absent both from lateral and midline glia (Fig. 3E-L), although Abd-A and Abd-B were expressed in a small subset of glia(Fig. 3L, arrowhead; not shown). We noticed that Hox expression boundaries were relatively imprecise,especially posteriorly; Hox expression domains did not end in well-defined lines across the cord, and some neurons lost Hox expression before others. The limits of Hox expression were also variable depending on the specific embryonic stage (Fig. 3Q; not shown). In the dMP2 neurons, Hox gene expression was particularly restricted;Ubx was absent from them in all segments(Fig. 3N,Q) and Antp was only expressed in the T2 dMP2 (Fig. 3M,Q). Expression of Abd-A spanned A5-A7(Fig. 3O,Q) and, importantly,Abd-B was specifically expressed in the A6-A8 dMP2s(Fig. 3P,Q). Thus, Hox gene expression in the Drosophila VNC is dynamic and complex, and Abd-B is the only Hox gene whose expression profile in dMP2 neurons parallels the segment specificity of their survival (Fig. 3Q).

Hox genes have been shown to activate apoptosis in post-embryonic neuroblasts and maxillary cells (Bello et al., 2003; Lohmann et al.,2002). However, the specific expression of Abd-B in posterior dMP2 neurons suggested that a Hox gene was involved in preventing neuronal apoptosis. To test this idea, we utilized dMP2-GAL4 to misexpress the four Hox genes in all dMP2 neurons postmitotically and cell-autonomously. Misexpression of either UAS-Antp or UAS-abd-A(Fig. 4A,C, respectively) had no effect, and UAS-Ubx resulted in only 8% rescue(Fig. 4B). In contrast, we observed nearly complete rescue of anterior dMP2s with UAS-Abd-B(79%; Fig. 4D). The Abd-B gene is transcribed in two forms, resulting in the expression of a `morphogenetic' (m) and a `regulatory' (r) protein that differ in their N-termini (Casanova et al.,1986; DeLorenzi et al.,1988; Zavortink and Sakonju,1989). We observed that both UAS-Abd-Bm (used above) and UAS-Abd-Br could efficiently suppress cell death in anterior dMP2s(76% survival for UAS-AbdBr, n=9 VNCs; not shown). The differential ability of these four Hox genes to prevent anterior dMP2 death was not caused by any obvious differences in their expression levels at the time when dMP2 neurons die. Indeed, at stage 16, we were able to detect robust ectopic expression of the four proteins in all dMP2s(Fig. 4E-H). When we used the pan-neuronal elav^GAL4 driver line to misexpress these four Hox genes, Abd-B prevented anterior dMP2 death completely(Fig. 6E,I, Fig. 4D for numbers), whereas expression of Antp, Ubx or abd-A increased dMP2 survival only marginally (Fig. 4A-D for numbers). The increased survival obtained with elav^GAL4 is likely to be a consequence of its earlier expression in the dMP2 lineage (not shown).

The surviving anterior dMP2 neurons expressed Odd(Fig. 4I) and appeared to differentiate fully, as evidenced by the expression of Proctolin and pMad(Fig. 4J; not shown). These terminal differentiation markers were also expressed in anterior dMP2s when their apoptosis was prevented using UAS-p35, or in cell death mutants(Fig. 2G,H). However, UAS-Abd-B expression appeared more potent in this respect, with 70%of `rescued' dMP2 cells expressing Proctolin in dMP2-GAL4/UAS-Abd-Bcompared to ∼30% in XR38/H99 mutants or dMP2-GAL4/UAS-p35. This indicates that in addition to its anti-apoptotic role in dMP2s Abd-B may also be involved in some aspects of dMP2 terminal differentiation.

The specific expression of Abd-B in posterior dMP2 neurons, as well as its ability to fully suppress anterior dMP2 cell death when misexpressed,suggested that Abd-B is normally involved in preventing apoptosis in posterior dMP2 neurons. In wild-type embryos, the two different Abd-B proteins are expressed in distinct domains; while Abd-Bm expression is apparent in segments A5 to A8, Abd-Br is confined to A9 (Boulet et al., 1991; Delorenzi and Bienz, 1990; Kuziora and McGinnis, 1988; Sanchez-Herrero and Crosby,1988) and should therefore not be expressed in posterior dMP2 neurons at the time of their death. To confirm this, and to determine whether Abd-Bm has an anti-apoptotic function in posterior dMP2 neurons, we analyzed the development of dMP2 neurons in Abd-Bm mutant embryos. In Abd-Bm mutants, dMP2 neurons were generated throughout the cord(including posterior segments) in normal numbers(Fig. 5A,B) and appeared properly specified, as evidenced by their expression of dMP2-GAL4 and Odd (Fig. 5F). As expected, at stage 16, Abd-B staining (corresponding to Abd-Br expression) was confined to the posterior tip of the VNC and was absent from posterior dMP2 neurons, including the posterior-most pair(Fig. 5E). At stage 17, all dMP2 neurons appeared to undergo apoptosis, and no posterior dMP2 neurons were detected at late embryonic stages using dMP2-GAL4 and Odd(Fig. 5D; Fig. 6D), strongly suggesting that Abd-Bm is normally required in posterior dMP2 neurons to prevent apoptosis. In contrast, and as would be expected from the Hox expression data(Fig. 3Q), the pattern of dMP2 survival/death was unaffected in mutant embryos lacking other Hox genes (not shown). To rule out a possible early contribution of Abd-Br to the generation of posterior dMP2s, we additionally analyzed Abd-B mutant embryos in which neither Abd-Bm nor Abd-Br is expressed, and obtained identical results to those described for Abd-Bm mutants (not shown). Thus, Abd-B is not required for the generation of posterior dMP2 neurons.

To confirm that the normal anti-apoptotic function of Abd-Bm is postmitotic and cell-autonomous, we attempted to rescue posterior dMP2 neurons by specifically expressing Abd-Bm using dMP2-GAL4 in an otherwise Abd-Bm mutant background. As expected, we observed robust rescue of posterior dMP2s (Fig. 5I), with numbers similar to those obtained for anterior dMP2s(Fig. 4D). Based on their expression of Odd, posterior dMP2s appeared to be properly specified(Fig. 5J).

The absence of rpr^GAL4 expression from posterior dMP2 neurons suggested that the segment specificity of dMP2 survival is achieved,at least in part, by the transcriptional repression of the rpr gene. Since Abd-B has been shown to activate rpr transcription in the periphery (Lohmann et al.,2002), it is conceivable that Abd-B can also act to repress rpr in certain cellular contexts. If this were the case in the dMP2 lineage, we would expect rpr^GAL4 to be expressed in posterior cells in Abd-B mutant embryos. To test this idea, we examined rpr^GAL4 expression in an Abd-Bm mutant background. While in control embryos no rpr^GAL4 expression was apparent in posterior dMP2 neurons(Fig. 2J), it was often observed in the posterior dMP2 neurons of Abd-Bm mutants(Fig. 5G).

Thus, in spite of its broad posterior expression, Abd-B is not necessary for the generation of the posterior dMP2 neurons. However, Abd-B is absolutely required cell-autonomously and postmitotically for the survival of posterior dMP2 neurons. Furthermore, the activation of rpr^GAL4 expression in posterior dMP2 neurons in Abd-B mutant embryos strongly suggests that Abd-B prevents dMP2 apoptosis, at least in part, by repressing the transcription of rpr.

In invertebrate embryos, developmental apoptosis typically takes place shortly after cells are generated. dMP2 cell death is unusual in this respect,in that it occurs at a later developmental stage when these neurons have differentiated and extended their axons, thus resembling apoptosis of developing vertebrate neurons. We wondered whether there were other cases of apoptosis of differentiated neurons in the VNC of the Drosophilaembryo, and whether Hox genes may be involved in regulating its segmental specificity.

Odd is expressed in both the dMP2 neurons and the smaller, more medial and more dorsal MP1 pioneer neurons (Fig. 1G,H). Previous studies have shown that the MP1 neuron is generated in all VNC segments and extends an ipsilateral projection that bifurcates both posteriorly and anteriorly(Bossing and Technau, 1994; Schmid et al., 1999). While using Odd as an additional marker for dMP2 neurons, we confirmed that MP1 neurons were present throughout the VNC at stage 16. However, we observed that Odd expression was conspicuously absent from anterior MP1 neurons in late embryos, and only A5-A8 MP1 neurons were apparent(Fig. 6A). Compared to dMP2 expression, Odd expression in MP1 neurons was lost at a slightly later stage(late stage 17, before tracheal inflation). Conceivably, this could have been a consequence of downregulation of Odd expression in anterior MP1s. However,while using Odd as a marker to identify dMP2 neurons in cell death mutants, we observed that its expression in anterior MP1 neurons was maintained in late embryos and larvae (H99/XR38; Fig. 6B); only two out of the four Odd-positive cells apparent in anterior segments were dMP2 neurons (Fig. 6H). This confirms that anterior MP1 neurons also undergo apoptosis at a late developmental stage.

Having identified an additional neuronal subtype that undergoes segment-specific apoptosis after differentiation, we sought to determine whether MP1 cell death was effected by the same molecular mechanisms involved in anterior dMP2 death. Similar to the complexity observed for dMP2 apoptosis,we found that several deletion combinations resulted in the survival of anterior MP1 neurons (Fig. 7B). Like dMP2 cell death, anterior MP1 death is mediated by both rpr and grim, while hid appears to play a very limited role. This is revealed by the almost complete rescue observed in XR38/H99 and X25/H99 mutants, but not in X14/H99 or X14/X14mutants (Fig. 6C; Fig. 7B). However, there is an important difference between dMP2 and MP1 death. In XR38/XR38 mutant embryos, a significant number of anterior dMP2 neurons survive but no MP1 neurons are rescued, indicating that loss of rpr alone cannot affect MP1 survival. By contrast, no anterior dMP2 neurons survive in X25/X25 (but not X14/X14) mutant embryos, whereas most MP1 neurons are rescued, implying that loss of grim alone is sufficient to prevent MP1 apoptosis (Fig. 6C). There is, however, some contribution of rpr to MP1 apoptosis since, in the absence of one copy of grim, loss of rpr is sufficient to rescue most MP1 neurons (XR38/H99, Fig. 6C). Thus, we conclude that grim and rpr effect apoptosis of both dMP2 and MP1 neurons. However, grim levels are critical in MP1 neurons, while dMP2 apoptosis depends more on rpr. The fact that we can readily detect rpr^GAL4 expression in anterior dMP2s, but only in nine anterior MP1 neurons out of 30 VNCs examined, supports this notion(Fig. 2J; Fig. 6G).

We next sought to determine whether Abd-B function underlies the posterior survival of MP1 neurons. We were initially surprised to find that in wild-type embryos MP1 neurons always survived one segment more anteriorly than dMP2 neurons (A5-A8; Fig. 6A). However, when we analyzed the expression of Abd-B in MP1 neurons, we found that not only was Abd-B expressed in posterior MP1 neurons, but its expression in this neuronal type extended into A5(Fig. 6F). Abd-B expression thus parallels the segmental survival of MP1 neurons. To test whether Abd-B functions in posterior MP1 neurons to prevent their death, we examined Abd-Bm mutants and found that MP1 neurons were initially generated throughout the VNC (Fig. 5F,asterisk). However, in late embryos, both anterior and posterior MP1 neurons were lost (Fig. 6D). Thus, Abd-B is required for the survival of posterior MP1 neurons, but not for their generation. Conversely (and as in the case of dMP2 neurons),anterior MP1 cell death could be completely suppressed by pan-neuronal expression of Abd-B (elav^GAL4/UAS-Abd-Bm; Fig. 6E).

The finding that the death of both anterior dMP2 and MP1 neurons is completely prevented by anterior expression of Abd-B cannot solely be explained by a repressive function of Abd-B on the rpr gene;loss of rpr does not result in survival of anterior MP1 neurons(XR38/XR38; Fig. 7B). Instead, MP1 survival requires an at least partial loss of grim(X25/X25 and XR38/H99; Fig. 7B). Furthermore, in dMP2 neurons, loss of rpr alone is not sufficient to prevent all anterior cell death; complete anterior rescue is only observed in mutants with additionally reduced grim gene dosage (XR38/XR38 versus XR38/H99; Fig. 7B). This implies either that Abd-B represses rpr and grim, or that it somehow acts downstream of both rpr and grim to prevent apoptosis in both lineages. To resolve this issue, we generated flies mutant for both the H99 region and Abd-Bm. In these double-mutants, dMP2 and MP1 neurons were properly specified throughout the cord, but they all failed to undergo apoptosis in late embryos(Fig. 8A,B). This indicates that loss of pro-apoptotic gene function is epistatic to Abd-Bfunction and probably places Abd-B upstream of the RHG-motif genes removed by the H99 deletion (Fig. 8C).

How does Abd-B prevent the function of RHG-motif genes? It is likely that Abd-B prevents pioneer neuron apoptosis by repressing the transcription of, at least, rpr and grim. This idea is supported by four facts: first, the H99 deletion is epistatic to Abd-B. Second, Abd-B is a transcription factor. Third, rpr^GAL4 is activated posteriorly in Abd-Bmmutants. Fourth, when misexpressed postmitotically, Abd-B can fully rescue both types of pioneer neurons. Given that loss of rpr is critical for anterior dMP2 survival, whereas loss of grim is critical for anterior MP1s, Abd-B must prevent the expression of at least these two cell death activators. This model is summarized in Fig. 8C.

In the developing vertebrate neural tube, a number of studies have shown that Hox genes are critical for AP organization and for proper neuronal specification (Capecchi, 1997; Carpenter, 2002; Lumsden and Krumlauf, 1996). Although their action may be largely confined to progenitor cells, recent studies have revealed that Hox genes can also act to control the identity of early postmitotic neurons (Dasen et al.,2003). In the light of our findings, it will be of interest to determine if selective, Hox-dependent elimination of mature neurons gives rise to differences in motor neuron numbers along the AP axis of the vertebrate spinal cord. Increased apoptosis of postmitotic motor neurons has been observed in mouse mutants lacking Hoxc-8, one of the vertebrate homologues of abd-A (Tiret et al., 1998). This may be the result of the aberrant connectivity pattern of Hoxc-8-deficient motor neurons, which would restrict their access to target-derived neurotrophic factors. However, this increase in cell death is also consistent with the possibility that Hoxc-8 normally acts to prevent apoptosis of postmitotic neurons in its expression domain.

Our results contrast with the previous finding that Abd-B appears to activate rpr transcription to regulate segment boundary formation in the posterior region of early Drosophila embryos(Lohmann et al., 2002). Decreased apoptosis has also been observed in mouse mutants lacking Hoxb13, one of the vertebrate homologues of Abd-B(Economides et al., 2003). It has previously been shown that the target functions of Hox genes are highly dependent on cellular context, and the regulation of apoptosis appears to be no exception. This context dependence may not be unique to the Abd-Bgene. abd-A has been previously reported to activate apoptosis in post-embryonic neuroblasts during normal development. When Antp and Ubx were misexpressed in these neuroblasts, they too were able to trigger apoptosis (Bello et al.,2003). In contrast, none of these genes acted in a pro-apoptotic manner in our study. It is, therefore, conceivable that the pro-apoptotic function of Hox genes is confined to progenitors, at least in the nervous system. Alternatively, or additionally, availability of certain cofactors may determine whether a Hox gene activates or represses transcription of pro-apoptotic genes in a specific cell.

In addition to their dependence on cellular context, specific Hox proteins may control pro-apoptotic genes differently. Abd-B and its vertebrate homologues share several properties that distinguish them from other Hox proteins, such as the absence of a hexapeptide motif and a preference for a different DNA core sequence (Pradel and White, 1998). Together, these differences may confer unique transcriptional properties on proteins of the Abd-B family, and may explain why Abd-B is the only Hox protein capable of fully rescuing anterior pioneer neurons. The finding that Abd-B is the only Hox gene that was unable to rescue the embryonic brain phenotypes of Drosophila mutants for the Hox gene labial is consistent with this idea(Hirth et al., 2001).

Our results show that while Hox genes are broadly expressed within their domains, they are largely absent from certain cell populations; at stage 16,few glial cells express Hox genes in the VNC. Since many Drosophilaneuroblasts give rise to both neurons and glia(Schmid et al., 1999; Schmidt et al., 1997), it is possible that Hox gene expression is actively suppressed by factors promoting glial fate. Alternatively, an initial wave of Hox expression in progenitors could be followed by a second, neuron-specific re-activation of Hox expression. In any case, it will be of interest to identify the molecular mechanism by which Hox gene expression is confined to specific populations of postmitotic cells in the nervous system.

While cellular context may determine whether a Hox gene acts in a pro- or anti-apoptotic manner, apoptosis of specific cells within a Hox expression domain may also be achieved by differential Hox gene expression. For example,while Abd-A is broadly expressed in abdominal segments during larval stages,it is absent from post-embryonic neuroblasts. However, at the last larval instar, a neuroblast-specific pulse of abd-A results in the activation of the cell death program in these cells(Bello et al., 2003). Similarly, and given the novel role for Hox proteins in the apoptosis (this study) and differentiation of postmitotic neurons(Dasen et al., 2003), the expression of Hox genes in specific postmitotic neurons is likely to be of functional significance. Together, these findings are not consistent with the view that Hox genes solely function as `segment identity' factors specifying global properties of the segments in which they are active. Instead, they lend functional support to the proposal that Hox genes are required for a number of decisions taken at the cellular level(Castelli-Gair, 1998; Hombria and Lovegrove,2003).

The combined activity of RHG-motif genes is critical to the initiation of all cell death in the Drosophila embryo(White et al., 1994). Both our findings and previous work indicate that these genes act in an additive manner(Zhou et al., 1997). However,not all cell death activators are simultaneously expressed in every cell fated to die, and their specific expression patterns do not always overlap(Chen et al., 1996; Grether et al., 1995; Robinow et al., 1997; White et al., 1994). Therefore, it is likely that they are differentially regulated by specific developmental signals. While Abd-B acts to repress rpr and grim function in posterior pioneer neurons, the developmental stimulus activating their expression in these neurons throughout the cord is currently unknown. Three developmental signals are known to regulate the function of RHG-motif genes in the Drosophila nervous system. The insect hormone ecdysone appears to be important for blocking cell death of certain peptidergic neurons during metamorphosis(Draizen et al., 1999). However, the ecdysone-receptor complex has also been shown to promote cell death by activating rpr transcription in other tissues during Drosophila metamorphosis (Jiang et al., 2000). While an embryonic ecdysone pulse occurs around the time when pioneer neurons die, our preliminary experiments have failed to lend any support to an ecdysone-dependent activation of apoptosis in these neurons(not shown). The EGF-receptor/Ras/MAPK pathway has been shown to phosphorylate Hid protein, thereby preventing apoptosis of midline glial cells(Bergmann et al., 2002). However, neither Rpr nor Grim appear to be regulated in this fashion, and this model would not address the specific transcriptional activation of these genes in pioneer neurons. Lastly, Notch signaling has been described as resulting in both activation and inhibition of apoptosis(Miele and Osborne, 1999). In Drosophila, recent studies have revealed that Notch can act cell-autonomously to induce apoptosis during final mitotic divisions both in the central and peripheral nervous systems(Lundell et al., 2003; Orgogozo et al., 2002). Although this Notch-induced developmental apoptosis is prevented in H99 mutant embryos, the molecular mechanisms by which activated Notch signaling results in the activation of IAP inhibitors are still unknown. Nevertheless, Notch signaling is unlikely to be relevant to dMP2 death, since it is not active in dMP2 neurons (Spana and Doe, 1996; Spana et al.,1995). It is, therefore, likely that an as yet unidentified factor is responsible for the activation of the apoptotic machinery in pioneer neurons. This factor could be Odd, given its specific expression in dMP2 and MP1 neurons. Because of the early role of odd in embryonic patterning(Nusslein-Volhard and Wieschaus,1980), its possible postmitotic function in these neurons cannot be addressed using the currently available odd mutants.

Developmental apoptosis in invertebrate embryos typically occurs shortly after cells are generated. In Drosophila, this has often precluded the identification of dying cells until apoptosis has been genetically prevented. Consequently, progress in identification of the mechanisms controlling apoptosis has been relatively slow, and little is known about the upstream pathways that initiate cell death in specific tissues or lineages. Furthermore, in the Drosophila VNC, studies have shown that apoptotic corpses are engulfed by glia, transported to the dorsal surface of the VNC and transferred to macrophages for final destruction(Freeman et al., 2003; Sears et al., 2003; Sonnenfeld and Jacobs, 1995). The molecular genetic mechanisms underlying this intriguing series of events are only just beginning to be unraveled(Baehrecke, 2002). The identification of a late apoptotic event in two of the best-studied and least complex lineages in the Drosophila CNS, as well as the characterization of the dMP2-GAL4 line, should contribute to the elucidation of the mechanisms involved in both the developmental initiation and execution of apoptosis.

We thank J. B. Thomas and V. Hartenstein for advice. We thank the Bloomington Stock Center, Flyview, Developmental Studies Hybridoma Bank, A. Brand, J. Castelli-Gair, I. Duncan, U. Gaul, F. Hirth, B. McGinnis, J. Nambu,D. Nässel, E. Sánchez-Herrero, J. Skeath, D. van Meyel, K. White and R. White for generously sharing fly lines, DNAs and antibodies. We are grateful to B. C. Bello, C. Q. Doe, F. Hirth, G. Morata, E. Sánchez-Herrero and J. B. Thomas for critically reading the manuscript. This work was supported by a grant from the Freudenberger Scholarship Fund at Harvard Medical School to S.T.