Targeted ablation of glia disrupts axon tract formation in the Drosophila CNS

In vertebrates glial cells appear, prior to neurite outgrowth, along the pathways that axons will subsequently follow. This observation led to the ‘blueprint hypothesis’ (Singer et al., 1979) which suggests that the paths growth cones recognise are formed by glial cells before axon outgrowth. Support for this hypothesis comes from experiments in which this preformed pathway is physically disrupted, leading to an accompanying disruption of the axon tract (Silver et al., 1982). More recent studies in the grasshopper and in Drosophila, by means of electron microscopy, inmunohistochemistry and cell lineage analysis, have demonstrated a similar organisation of glial cells along axonal pathways preceding extension of growth cones (Fredieu and Mahowald, 1989; Jacobs and Goodman, 1989a; Jacobs et al., 1989; Klämbt and Goodman, 1991; Goodman and Doe, 1993). Furthermore, Jacobs and Goodman (1989a) proposed that molecular heterogeneities within the glial population might allow pioneer growth cones to distinguish between different pathways.

The first evidence in support of glial cells acting as guideposts in insects was provided by experiments in grasshopper in which one glial cell type, the segment boundary cell (SBC), was physically ablated using a laser (Bastiani and Goodman, 1986). In grasshopper embryos the U axons, which pioneer the intersegmental nerve, contact the SBC prior to exiting the central nervous system (CNS). The growth cone of the aCC motor neuron follows the U axons to exit the ventral nerve cord in the intersegmental nerve. After ablation of the SBC, Bastiani and Goodman (1986) labelled the aCC neuron and showed that the aCC growth cone continues past its normal exit point from the CNS, implying that the U axons were unable to exit the CNS.

Here we investigate the role that glial cells play in guiding growth cones during the formation of the axon tracts within the Drosophila embryonic CNS. The ventral nerve cord of the CNS is characterised by two longitudinal bundles of axons (the longitudinal connectives) that run along the length of the embryo. These are joined in each segment, in a ladder-like fashion, by two commissural fascicles that cross the midline (reviewed by Goodman and Doe, 1993).

Prior to axon outgrowth, the axon scaffold is prefigured by an array of glial cells. In each hemisegment the longitudinal glia originate from a lateral precursor cell, or glioblast, which can first be detected at stage 11 (Jacobs and Goodman, 1989a; Jacobs et al., 1989; Campbell et al., 1994; Xiong et al., 1994; Halter et al., 1995; Ito et al., 1995; stages according to CamposOrtega and Hartenstein, 1985). The glioblast divides once symmetrically and the two glial progeny migrate along the dorsal surface of the developing nerve cord, in an anterior and ventral direction towards the midline (Fig. 1). The progeny of the glioblast then divide and intermingle with two additional glial cells to give rise to a group of six cells collectively named the interface glia (see Ito et al., 1995 for a description of the embryonic glia). By the middle of stage 12, these glia prefigure the longitudinal axon tracts that will extend along the length of the embryo. Later in development some of these cells will divide again resulting in two rows of 9-10 cells per hemisegment overlying the longitudinal axon tracts.

By means of electron microscopy and immunohistochemistry, studies in grasshopper and Drosophila embryos have shown that the growth cones of the MP1, vMP2, dMP2 and pCC neurons, which pioneer the first two longitudinal tracts, make contacts with some of these glial cells during their outgrowth (Jacobs and Goodman, 1989a,b; Lin et al., 1994). This suggests that at least some of the interface glial cells might be guidepost cells which play a fundamental role in establishing the longitudinal tracts.

Several mutants that show severe malformation of the longitudinal tracts also exhibit disruptions in the longitudinal glia (Doe et al., 1991; Jacobs, 1993; Klaes et al., 1994). While this observation would seem to support the hypothesis that glia act as guidepost cells, most of the mutants so far investigated are not glial-specific. For example, the neuropile phenotype seen in longitudinals lacking, orthodenticle and prospero mutants (Finkelstein et al., 1990; Doe et al., 1991; Seeger et al., 1993; Giniger et al., 1994) could be due to the misdifferentiation of outgrowing neurons, as these genes are expressed not only in the glia, but also in other cells in the CNS.

The expression of two genes, reversed polarity (repo) and pointed, is restricted to glial cells (Klämbt, 1993; Campbell et al., 1994; Klaes et al., 1994; Xiong et al., 1994; Halter et al., 1995). Both genes encode putative transcription factors that are required for glial differentiation. Mutations in pointed affect axon outgrowth: the longitudinal connectives are occasionally disrupted and the commissural axon tracts fuse. Since pointed is expressed in many types of glia, including the longitudinal glia, the VUM support glia (or MM-CBG in Ito et al., 1995) and the midline glia (Klämbt, 1993), the pointed phenotype lends support to the hypothesis that glial cells might play an important role in axonal pathfinding. In repo mutant embryos, however, the axon tracts are able to form. The defects that arise, which include weakening of the axon fascicles, are only seen later in development. The repo phenotype might therefore suggest a role for glia in the maintenance, rather than in the establishment, of the fiber tracts.

Cell ablation provides a means of directly testing the function of specific glial cells. We have developed a method of genetic cell ablation (A. H. B., C. Davidson, K. Yoffe, C. O’Kane, A. H. and K. Moffat, unpublished data), by targeting expression of a toxin with the GAL4 system (Brand and Perrimon, 1993; Brand et al., 1994). This method does not require the use of lasers or any other invasive manipulation, and can be used to kill a particular cell type in every segment of the embryo. After a simple genetic cross, it is possible to collect and analyse hundreds of embryos in each experiment. We have used toxin ablation to kill the interface glia at different stages of embryonic development, and show here that the glia are essential for the formation of the longitudinal tracts, and may also play a role in the maintenance of the longitudinal connectives.

Flies were raised on standard Drosophila media at 25°C, unless otherwise noted. Descriptions of balancers and mutations that are not described in the text can be found in Lindsley and Zimm (1992). Embryonic stages are as described by Campos-Ortega and Hartenstein (1985).

Lines that express GAL4 in a restricted cellor tissue-specific pattern were generated by enhancer detection (Brand and Perrimon, 1993). An enhancerless gene encoding the yeast transcriptional activator GAL4 is inserted randomly into the Drosophila genome where, depending upon the site of integration, expression is directed by any one of a diverse array of genomic enhancers. A second gene, containing GAL4-binding sites (or Upstream Activation Sequence, UAS) within its promoter, can then be introduced into this background where it will only be transcribed in those cells where GAL4 is expressed. Each GAL4 insertion line was crossed initially to a line carrying the UAS-lacZ reporter gene (UAS-lacZ^4-1-2; Brand and Perrimon, 1993). Embryos from the cross were stained for β-galactosidase expression with anti-β-galactosidase antibodies. Lines of interest were subsequently rescreened using the UAS-tau-lacZ insertion line described below.

The tau-lacZ fusion gene, including an SV40 terminator, was excised as an XhoI-XbaI fragment from plasmid tau-lacZ/pBS KS+ (a gift from C. Callahan and J. Thomas), and subcloned into vector pUAST (Brand and Perrimon, 1993) to give UAS-tau-lacZ. Transgenic lines were generated by injection of DNA, prepared following the Qiagen midiprep protocol, at a concentration of 600 μg/ml, into embryos of strain y w; Sb, P[ry+, Δ2-3]/TM6 (Robertson et al., 1988) using standard procedures (Spradling, 1986). Several independent transformants were obtained.

The construction of the transgenic line carrying the UAS-ricin A gene will be described elsewhere (A. H. B., C. Davidson, K. Yoffe, C. O’Kane, A. H. and K. Moffat, unpublished data). In brief, to prevent transient expression of the toxin following injection, a copy of the white gene was inserted between the GAL4 UAS and the ricin A coding sequence. Flp recombinase target sites (FRTs) flank the white gene and, following transformation, white can be excised by introduction of flp recombinase in a genetic cross. UAS-ricin A is a lethal insertion and is kept as a stock balanced over CyO. The balancer chromosome carries an insertion of an en-lacZ gene (Perrimon et al., 1991) at the wingless locus (CyO, wg^en11; Perrimon et al., unpublished). When GAL4 lines are crossed to UAS-ricin A/CyO, wg^en11, embryos carrying the balancer chromosome can be identified by staining with anti-β-galactosidase antibodies, revealing a wingless expression pattern.

Antibody staining reactions were carried out by standard methods (Patel, 1994). Antibodies were diluted in PBS, 0.1% Triton X-100, and used at the following concentrations: rabbit anti-β-galactosidase 1:10000 (Cappel); mouse anti-β-galactosidase at 1:100 (Promega); rabbit anti-repo at 1:300 (Halter et al., 1995; a gift from D. Halter); mouse mAb BP102 at 1:5-1:10 (Seeger et al., 1993; Patel, 1994; a gift from N. Patel); mouse mAb 1D4 anti-fasciclin II at 1:8 (Van Vactor et al., 1993; a gift from C. Goodman and K. Broadie); mouse mAb 22C10 at 1:50 (Fujita et al., 1982; a gift from M. Ruiz-Gomez); mouse anti-pros at 1:100 (Doe et al., 1991; a gift from S. Romani). Secondary antibodies (1:300) were either directly conjugated to FITC, Texas Red, or horseradish peroxidase (HRP) (Jackson Labs), or were biotinylated (Vector Labs) and detected after incubation for half an hour in avidin and biotinylated HRP (Vectastain Elite ABC Kit, Vector Laboratories). Peroxidase was detected using diaminobenzidine (0.3 mg/ml in PBS, 0.1% Triton X-100; Sigma) as a substrate. For double labelling experiments with HRP, antibodies were incubated sequentially and the first staining was intensified with 0.08% NiCl. Embryos were cleared overnight in 50% glycerol, 0.1% Triton X-100 in PBS and were then transferred to 70% glycerol, 0.1% Triton X-100 in PBS. After antibody staining, embryos were prepared either as whole mounts, or ventral nerve cords were dissected using tungsten needles. Transmitted light images were generated by DIC microscopy using a Zeiss Axiophot.

For fluorescence detection, antibody concentrations were doubled and embryos were mounted in Vectashield (Vector Labs) and kept at 4°C. Confocal images are projections of ten to fifteen 2 μm optical sections through the ventral nerve cord, generated using a Biorad MRC600 confocal microscope.

Images were imported into Adobe Photoshop, assembled using Adobe Illustrator and printed on a Tektronix Phaser 440 laser printer.

To specifically ablate the GAL4-expressing cells, each GAL4 insertion line was crossed to a second transgenic line carrying a GAL4-responsive toxin gene, UAS-ricin A (A. H. B., C. Davidson, K. Yoffe, C. O’Kane, A. H. and K. Moffat, unpublished data). Ricin consists of two polypeptide chains: the B chain binds to the cell surface and allows the toxin to be internalised; the A chain inactivates eukaryotic ribosomes through depurination of 28S ribosomal RNA, which irreversibly inhibits protein translation (Endo and Tsurugi, 1988). The GAL4-responsive toxin gene, UAS-ricin A, encodes the catalytic subunit of the toxin only (Moffat et al., 1992). Since the B chain, which enables the toxin to cross the cell membrane, has been removed, toxin-induced ablation should be cell autonomous: the toxin will be unable to enter and kill neighbouring cells.

Compelling evidence for the efficiency and specificity of cell ablation is provided by experiments in which cells in the peripheral nervous system are killed. In line 12M, GAL4 is expressed not only in the interface glia in the CNS, but also in each of the four clusters of cells in each segment that make up the PNS (Fig. 2A,D; Ghysen et al., 1986). In particular, GAL4 drives expression of tau-β-galactosidase in five of the nine cells in the v’ cluster and in all but one of the cells of the lateral cluster, including the five chordotonal neurons (Fig. 2A,D). After these cells are killed by expression of ricin A, four cells from the v’ cluster (including the cell that migrates to join the lateral cluster) and one cell in the lateral cluster remain (Fig. 2C,F). In the v’ cluster the immediate neighbours of the toxinablated cells survive. Of the ten to twelve cells in the d cluster (Fig. 2B,E), three to four express GAL4 (Fig. 2A,D). After ablation, eight or nine cells remain in this cluster (Fig. 2C,F). Toxin ablation appears to be extremely efficient and reproducible throughout the embryo, but is cell-specific: the toxin kills those cells in which it is expressed but is not taken up by neighbouring cells.

We have used four independent lines that express GAL4 in the interface glia to target expression of the toxin, ricin-A (Fig. 3). To reveal the embryonic GAL4 expression pattern, each GAL4 line was crossed to a transgenic line carrying a UAS-tau-lacZ reporter gene (see Materials and Methods) to drive expression of β-galactosidase in the GAL4-expressing cells. Tau is a microtubule-associated protein (MAP) which, when fused to β-galactosidase to create a microtubule-bound reporter protein, permits the entire cell to be visualised (Butner and Kirschner, 1991; Callahan and Thomas, 1994). Neuronal cell bodies and their axons are easily identified, as are glial cells and their projections. To confirm that the cells expressing GAL4 are glia, we stained embryos expressing tau-β galactosidase with antibodies raised against the glial-specific antigen repo. repo is a homeodomain DNA-binding protein that is found in the nucleus of all glial cells except the midline glia (Campbell et al., 1994; Xiong et al., 1994; Halter et al., 1995). The GAL4expressing cells identified as glia based on their position and morphology also express repo (Fig. 4).

The four GAL4 lines selected allow us to kill the longitudinal glia at different stages of embryonic development. Line MZ1580 expresses GAL4 from stage 11 in the longitudinal glioblast and its progeny (Figs 3A, 4A), and later in most other glial cells, in the MP2 neurons and in macrophages (Fig. 3B,C; see below). Ablation of the GAL4-expressing cells in line MZ1580, by activation of UAS-ricin A, eliminates the longitudinal tracts (Fig. 5B). Although the longitudinal connectives are absent, the commissures remain intact. The commissural fascicles appear to be thicker, suggesting that the longitudinal axons may be misrouted across the midline. The loss of the longitudinal connectives is a highly penetrant phenotype, with 71% of the embryos scored showing disruptions (Table 1). The expressivity of the phenotype, as measured by counting the number of hemisegments showing longitudinal defects (Table 2), is also high. When scoring embryos stained with mAb BP102 (Patel, 1994; Seeger et al., 1993), which labels axons in the CNS, the longitudinal connectives are disrupted in 211/292, or 72%, of hemisegments. In embryos stained with anti-fasciclin II, which labels the surface of the neurons that pioneer the longitudinal axon tracts (pCC, vMP2, dMP2 and MP1; Lin et al., 1994; Van Vactor et al., 1993) 309/330, or 93%, of hemisegments show defects.

To confirm that the interface glia have been killed, embryos were stained with anti-repo antibodies (Campbell et al., 1994; Xiong et al., 1994; Halter et al., 1995). In embryos expressing ricin A, virtually all of the repo-positive glial cells are missing (Fig. 5B). The number of repo-positive glia overlying the neuropile is reduced from an average of 11.5 cells per hemisegment in wild-type embryos (Oregon P2; 28 hemisegments counted) to 2.6 cells per hemisegment when the GAL4-positive cells are killed (82 hemisegments counted).

In line MZ1580, GAL4 is also expressed in other glial cells: the A and B glia and other subperineurial glia, the cell body glia, the dorsal and ventral channel glia, the intersegmental nerve root glia (including the segment boundary cell) and the exit glia (see Ito et al., 1995 for a description of the embryonic glia). Most of these glia do not lie over the longitudinal axon tracts, however, and are therefore unlikely to be required for longitudinal axon outgrowth. GAL4 is also expressed in the MP2 neurons, which help to pioneer the longitudinal axon tracts and in macrophages. Although the loss of these pioneer neurons might be predicted to disrupt axon tract formation, we rarely see breaks in the longitudinal tracts when the MP2 neurons alone are ablated (10% of embryos scored; A. H. and A. H. B., unpublished data). The array of glial cells that prefigures the axon scaffold appears, therefore, to be essential for the establishment of the longitudinal axon tracts.

Lines C321c, 12M and MZ1131 express GAL4 in the interface glia, albeit later in development than line MZ1580 (Figs 3, 4). For example, none of these lines express GAL4 in the longitudinal glioblast. Line C321c starts to express GAL4 in the interface glia in a small number of hemisegments at stage 13 and more extensively from stage 14, after the glia have migrated ventrally (Fig. 3D). Line 12M expresses GAL4 in the interface glia from stage 14 (Fig. 3 G), and line MZ1131 from stage 13 (Fig. 3J). Since the longitudinal connectives are pioneered at the end of stage 12, these lines should drive expression of ricin A in the glia after the pioneer growth cones have started to grow out.

Ablation of the GAL4-expressing cells in line C321c diminishes the longitudinal connectives (Fig. 5C,D; Table 1 and 2). The loss of axons is associated with a reduction in the number of glial cells along the axon tracts, as detected by staining with repo-specific antibodies (Fig. 5C,D). The glial cells that remain are disorganised and tend to cluster over the commissures, where the axon fascicles can be considerably broader than in wild-type embryos. This might be due to misrouting of axons that would normally grow along the longitudinal tracts.

The longitudinal connectives are also reduced or interrupted after killing the GAL4-expressing cells in line 12M (Fig. 5E) and line MZ1131 (Fig. 5F). The thin and broken tracts correspond with regions of the neuropile where there are fewer glial cells. By labelling embryos with antibodies that recognize prospero (Doe et al., 1991), we can show that glia derived from the longitudinal glioblast are missing . At stage 16, prospero is expressed in the five or six progeny of the glioblast (Fig. 6B). After the GAL4-expressing cells have been ablated in line MZ1131, only two or three prospero-expressing cells remain per hemisegment (Fig. 6D,E).

Since GAL4 expression in the CNS of lines C321c, 12M and MZ1131 is restricted to the interface glia, we conclude that the loss of these glia alone is sufficient to disrupt the longitudinal tracts. Killing the glia after the first fascicles have formed still disrupts the axon tracts, suggesting that the interface glia are required first for pioneer axon outgrowth and then either to direct the follower growth cones, or to maintain the longitudinal axon tracts.

We followed the effect of glial ablation at different stages of development by staining embryos with antibodies raised against fasciclin II, which is expressed on the surface of the neurons that pioneer the longitudinal connectives. The first longitudinal tract is pioneered towards the end of stage 12 by the ascending growth cones of the pCC and vMP2 neurons, which meet the descending growth cones of dMP2 and MP1 (Jacobs and Goodman, 1989a,b; Lin et al., 1994) (Fig. 7A). At stage 16, fasciclin II expression defines two longitudinal axon bundles per hemisegment (Fig. 7C), and by stage 17, embryos express fasciclin II in three distinct fascicles per hemisegment (Fig. 7E).

When using line MZ1580 to express ricin, the longitudinal glial precursor and the interface glia that arise from other lineages are killed, and the glial scaffold does not form prior to axon outgrowth (Fig. 6C). As a result, the first axon fascicle is not formed (Fig. 6C) and the longitudinal connectives are disrupted in 93% of hemisegments (as detected by fasciclin II expression; Table 2). Using lines C321c, 12M and MZ1131, the glia are killed later in development, after the first growth cones have started to pioneer the longitudinal tracts, and the connectives in 40 to 60% of hemisegments are thin or broken (Figs 5, 6D,E, 7; Table 2).

We followed axon tract formation in embryos in which the glia are killed when line C321c drives expression of UAS-ricin A. The first longitudinal tract is established in wild-type embryos by stage 13 (Fig. 7A), when GAL4 is initially expressed in C321c. At this stage some gaps are observed in the longitudinal connectives in C321c UAS-ricin A embryos (Fig. 7B), suggesting that the toxin is active virtually immediately following GAL4 expression. These defects become more frequent at stage 14, when C321c expresses GAL4 more extensively in the interface glia. At stage 16, gaps can be seen in one or both of the longitudinal axon bundles and axons are often misrouted (Fig. 7D). By stage 17, ablation of the interface glia causes irregularities and breaks in at least one of the axon fascicles (Fig. 7F).

It has been proposed that one of the roles played by glia is to serve as guideposts for the growth of pioneering axons. We tested this by genetically ablating glial cells and monitoring axon outgrowth in their absence. To kill specific cells in the nervous system, we made use of a technique in which GAL4 drives the expression of the toxin ricin A (A. H. B., C. Davidson, K. Yoffe, C. O’Kane, A. H. and K. Moffat, unpublished data). This is a simple and efficient method of targeted cell ablation, that has several advantages over other ablation systems. First, cell killing is rapid because the wild-type toxin is expressed; temperature-sensitive toxins (Bellen et al., 1992; Moffat et al., 1992) take much longer to kill cells. We have been unable to generate reproducible embryonic phenotypes using GAL4 to drive expression of either cold-sensitive ricin or heat-sensitive diphtheria toxin (A.B., unpublished). Second, no invasive manipulation of the embryos is required. Two lines are crossed, one expressing GAL4 in the cells of interest and the second carrying the UAS-ricin A gene. In the progeny of the cross, the GAL4-expressing cells are killed in each segment of every embryo that carries both transgenes. No manipulation of the embryos is necessary in order to induce cell killing. This contrasts with a recently described technique for embryonic cell ablation (Lin et al., 1995) in which each embryo must be injected with RNA. Third, cell ablation is autonomous. We have shown that when cells are killed by toxin expresssion, their immediate neighbours survive. Non-autonomous effects seen after toxin ablation are, therefore, likely due to the loss of the toxin ablated cells, rather than to non-specific killing resulting from toxin leakage.

We have used GAL4-mediated ablation to kill the interface glia (see nomenclature in Ito et al., 1995) that prefigure and then overlie the longitudinal axon tracts of the embryonic ventral nerve cord. Four independent GAL4-expressing lines were used in these experiments: one drives expression in the longitudinal glioblast and in its progeny, and three express GAL4 in the interface glia once they have reached their medial-most position in the nerve cord. Although these lines differ in their patterns of GAL4 expression, they all express GAL4 in the interface glial cells. Ablation of the GAL4expressing cells in all four lines leads to the misrouting, thinning or elimination of the longitudinal axon tracts. Since the interface glia are ablated in all four lines we conclude that these glial cells are required for the formation of the longitudinal connectives.

The most severe phenotype, with the highest penetrance and expressivity after cell ablation, is observed using line MZ1580, which expresses GAL4 in most of the embryonic glia and in the MP2 neurons. This is the only line that we have found to date that expresses GAL4 from stage 11 in the longitudinal glioblast, the progenitor cell that gives rise to most of the interface glia. It has recently been reported that the neurons that pioneer the longitudinal connectives can be killed without dramatically affecting the longitudinal axon tracts (Lin et al., 1995). We might infer from these results that the MZ1580 phenotype is due only to the loss of glial cells. However, we find that killing the MP2 neurons alone does affect axon tract formation (A. H. and A. H. B., unpublished data), although the phenotype is much less severe than when line MZ1580 drives toxin expression. The MZ1580 phenotype seems, therefore, to be due primarily to the early ablation of the longitudinal glia. However, we cannot rule out the possibility that the loss of the MP2 neurons contributes, either additively or synergistically, to the phenotype.

Interestingly, ablation of the interface glia at later stages of development (such as stage 13/14, with lines C321c, 12M and MZ1131) also leads to defects in the longitudinal tracts. This argues that the interface glia are necessary for longitudinal tract formation even after the axon scaffold has been initiated by the pioneer growth cones (at stage 12/2). The glia might provide positional cues not only for the pioneer growth cones but also for the following axons. Alternatively, the glia might be necessary to maintain the axon scaffold established by the pioneer neurons, such that the scaffold disintegrates as a consequence of glial cell death.

The GAL4 system has allowed us to ablate a selected population of glial cells in the Drosophila embryonic central nervous system. Targeted cell killing is a direct means of testing the role of cell-cell interaction in development. We have demonstrated that toxin ablation is autonomous, such that nonautonomous effects can be attributed to the loss of glialspecific signals. We have shown that the longitudinal glia are essential for the formation of the longitudinal axon tracts. These observations are consistent with a role for these cells as guide posts for the outgrowth of pioneer growth cones. The phenotypes that we observe when killing glial cells later in embryonic development, after growth cones have pioneered the first axon tracts, suggest that the interface glia may also play a role in the guidance of the axons that follow the pioneer axons, and/or in maintaining the axon scaffold once it has been established.

We thank Gerd Technau for helpful discussions throughout the course of this work. We are most grateful to Chris Callahan and John Thomas for making the tau-lacZ gene available to us prior to publication, to Gerd Technau, in whose laboratory some of this work was carried out, to Kim Kaiser for line C321c, and to Norbert Perrimon for help in making line 12M and for making available CyO,wg^en11. For antibodies we would like to thank Nipam Patel, Corey Goodman, Daniel Halter, Seymour Benzer, Mar Ruiz Gomez, Susana Romani and Kendal Broadie. Many thanks to Gerd Technau, Michael Bate, Lola Martin Bermudo, Daniel St. Johnston, Michalis Averof and Ulrik John for comments on the manuscript. This work was funded by a Wellcome Trust Senior Fellowship to A. H. B. and by DFG grant Te130/4-1 and EEC grant SCICT92-0790 to J. U.