β-Spectrin functions independently of Ankyrin to regulate the establishment and maintenance of axon connections in the Drosophilaembryonic CNS

During central nervous system (CNS) development, growth cones navigate a series of choice points to find their appropriate targets. These guidance decisions are shaped by a balance of attractive and repulsive cues found in the extracellular environment that can act locally or at a distance(Tessier-Lavigne and Goodman,1996). Slit ligands and Robo receptors play conserved roles in regulating midline axon crossing in both invertebrate and vertebrate nervous systems. In the Drosophila embryonic CNS, ipsilateral neurons(neurons that do not cross the midline) express the Robo receptor on their surface, which senses Slit to prevent midline crossing(Kidd et al., 1999; Kidd et al., 1998). By contrast, commissural neurons can overcome Slit repulsion because expression of Robo on their surface is prevented prior to midline crossing by the Commissureless (Comm) protein (Keleman et al., 2002). Thus, the specific receptors that a growth cone expresses as it encounters the midline greatly influences its guidance decision. The question of how guidance receptors and their downstream effectors are targeted to and distributed within functional domains of the growth cone plasma membrane remains key to the understanding of the mechanisms of axon path finding.

The Spectrin molecule, a long rod-shaped heterotetramer consisting of twoα and two β subunits, is the defining element of a ubiquitous sub-membrane cytoskeletal network in nearly all metazoan cells(Bennett and Baines, 2001). Most β-Spectrin isoforms contain an actin-binding domain at their N termini, an Ankyrin-binding domain, and a pleckstrin homology (PH) domain near their C termini. The PH domain of β-Spectrin interacts with the plasma membrane in transfected COS cells (Wang et al., 1996). α-Spectrin contains an SH3 domain and two EF-hand motifs (Bennett and Baines,2001).

In red blood cells, the Spectrin network was originally shown to have a role in supporting cell shape (Bennett and Chen, 2001); however, more recently it has been shown to participate in the formation of specialized membrane sub-domains(Bennett and Chen, 2001). Through its binding partner Ankyrin, the Spectrin network links many integral membrane proteins to the actin cytoskeleton. Genetic studies have revealed that many membrane proteins are mislocalized in the absence of Spectrin or Ankyrin (Dubreuil et al.,2000; Jenkins and Bennett,2001; Komada and Soriano,2002; Zhou et al.,1998). In addition, mutations in the vertebrate β-Spectrin and Ankyrin genes have been linked to various diseases, such as hereditary hemolytic anemias and spinocerebellar ataxia in humans, and auditory and motor neuropathies in mice (Ikeda et al.,2006; Mohler and Bennett,2005; Parkinson et al.,2001). It has been proposed that Ankyrin and β-Spectrin are interdependent, and mutually stabilize the formation of polarized domains at axon initial segments and nodes of Ranvier(Komada and Soriano, 2002). Yet the relationship between Ankyrin and β-Spectrin, and the role that each plays in coordinately regulating microdomain assembly, is not well understood.

In Drosophila, β-spectrin mutants die before larval hatching and loss of β-Spectrin leads to a failure of Na,K ATPase accumulation at the basolateral domain of midgut epithelial cells(Dubreuil et al., 2000). More recently, studies using RNA interference (RNAi) techniques have demonstrated a role for β-Spectrin during nervous system development. Specifically, theβ-Spectrin network was shown to be required pre-synaptically to maintain cell adhesion molecule (CAM) organization and neuromuscular junction stability(Pielage et al., 2005).

In a genetic screen for additional genes involved in midline axon guidance in the Drosophila CNS, we identified three β-spectrinalleles. β-spectrin mutant embryos display fused anterior and posterior commissures, and ectopic Fas2-positive neurons crossing the midline,suggesting a role for Spectrin in midline repulsion. Likewise,α -spectrin mutants exhibit very mild defects in midline repulsion, supporting the idea that it too has a minor role in regulating guidance. Additionally, in α-spectrin mutants, β-Spectrin is preferentially distributed to CNS axons, suggesting that α-Spectrin normally limits axonal accumulation of β-Spectrin. Genetic rescue experiments demonstrate that β-Spectrin functions in neurons and that the Ankyrin-binding domain is not required for guidance. This was surprising as Ankyrin is thought to be intimately associated with Spectrin in most cases. Furthermore, β-spectrin mutants demonstrate dose-dependent genetic interactions with the repulsive ligand slit and its receptor robo (but not with other guidance molecules), suggesting that they may be involved in the same pathway. Similar genetic interactions are observed within a small subpopulation of ipsilateral neurons, the apterous (Ap)neurons. Defects in the Ap neurons appear later in development, suggesting a role for β-spectrin in the maintenance of proper connectivity. Our data support a model in which the Spectrin network is important for the sorting and/or localization of molecules that contribute to and maintain midline repulsion.

Hülsmeier et al. (Hülsmeier et al., 2007) have reported that β-spectrin is the gene disrupted in the karussel mutant; karussel was previously isolated in a large-scale genetic screen for defects in axon guidance at the CNS midline (Hummel et al., 1999b). Our results, together with this complementary study,strongly support an important role for β-spectrin in the regulation of connectivity in the embryonic CNS.

The following stocks were used: (1)βspec^em6/FM7actinβgal; (2)βspec^em21/FM7actinβgal; (3)βspec^G0108/FM7actinβgal; (4)βspec^G0198/FM7actinβgal; (5)βspec^G0074/FM7actinβgal; (6)αSpec^rg41/Tm3UbxβGal; (7)αSpec^lm88/Tm3UbxβGal; (8)αSpec^lm32/Tm3UbxβGal; (9) slit²,robo^GA285/CyOWgβgal; (10) slit²,robo^GA285/CyOWgβgal; elavGAL4; (11) slit², aptGAL4,UAS-TauMycGFP/CyOTubulinGal80; (12) aptGAL4, UAS-TauMycGFP/CyOTubulinGal80;(13) robo^GA285, aptGAL4/CyOWgβgal; (14)robo^GA285/CyOWgβgal; UAS-TauGFP; (15) yw^iso3; (16)elavGAL4 (III); and (17) single-mindedGAL4 (simGAL4).

To identify genes involved in midline guidance, we screened for defects in the eagle neurons, a subset of commissural neurons that require frazzled for midline attraction (D.S.G. and G.J.B., unpublished)using a reporter stock: fra, UAS-TauMycGFP/CyTubulinGal80; eagleGAL4. Reporter males were crossed to a collection of lethal p-element lines on the X chromosome. Embryos were fixed and stained using a `large-scale collection'technique modified from Hummel et al.(Hummel et al., 1997). General axon guidance defects were identified using anti-BP102 and subset guidance errors were identified using anti-GFP.

βspec^em6/FM7actinβgal;UASβSpectrin-Myc flies were generated and crossed to elavGAL4 or simGAL4 males.

βSpec^hypo/FM7actinβgal;CxD/TM3Ubxβgal virgins were crossed toα Spec^lm88 or ^rg41/TM3Ubxβgal males. Non-FM7, non-CxD virgins (βSpec/+; αSpec/TM3Ubxβgal)were crossed back to αSpec/TM3Ubxβgal males. Hemizygous β-spectrin and homozygous α-spectrinmutant embryos were identified by the absence of staining for Sex Lethal(Sxl), β-Spectrin or β-gal.

Male flies carrying ubiquitously expressed wild-type (KW3A) or mutant transgenes (βSpecΔank or βSpecΔPH) were crossed toβSpec^em6/FM7actinβgal virgins. Embryos negative for Sxl and β-gal but positive for Myc were scored for rescue using an anti-1D4/Fas2 antibody.

The following primary antibodies were used: (1) mouse anti-1D4/Fas2[Developmental Studies Hybridoma Bank (DSHB), 1:100]; (2) mouse anti-BP102(DSHB, 1:100); (3) rabbit anti-Myc (Sigma-Aldrich, 1:500); (4) rabbit anti-β-Spectrin [337 (Byers et al.,1989; Dubreuil et al.,2000)]; (5) mouse anti-α-Spectrin (DSHB, 3A9, 1:25); (6)mouse anti-Sex Lethal (DSHB, M18, 1:1000); (7) mouse anti-β-gal (DSHB,40-1a, 1:250); (8) Rb-anti-GFP (Molecular Probes, 1:500); (9) mouse anti-Robo(DSHB, 1:50). The following secondary antibodies were used: AlexaFluor488 goat anti-Rb (Molecular Probes, 1:500) and Cy3-conjugated goat anti-mouse (Jackson Laboratories, 1:1000). Stacks of images were obtained using a Leica DMIRE2 confocal and a 63× oil immersion objective. Stacks were generated using NIH ImageJ software.

We confirmed the online p-element insert sequence annotation(http://flybase.bio.indiana.edu/)by recovering and sequencing the genomic DNA flanking the p-elements using standard protocols and primers (Huang et al., 2000). Although the p-element insertion sites associated with alleles G0108 and G0074 are only approximately 60 nucleotides apart from each other (Fig. 1A), G0108 embryos display a much more severe defect than do G0074 embryos.

Previous genetic screens looking for defects in CNS architecture have been instrumental in our understanding of axon guidance(Hummel et al., 1999a; Hummel et al., 1999b; Seeger, 1994). In an attempt to identify additional axon guidance molecules, we screened for global axon guidance defects and also more specifically for axon guidance errors within a subset of commissural neurons, the eagle neurons, labeled with GFP (see Materials and methods). We screened a collection of lethal p-elements on the X chromosome and identified a p-element line (0011819, allele G0108),in which hemizygous males exhibit a compressed axon scaffold containing fused anterior and posterior commissures. Public database(http://flybase.bio.indiana.edu)annotation indicates that the p-element disrupts the β-spectrinlocus (Fig. 1A). There were two additional β-spectrin alleles(Fig. 1A) contained in the p-element collection (line 11799, allele G0074; and line 11922,allele G0198), both of which display phenotypes similar to the originally identified allele, although less severe(Fig. 1I). A strong loss-of-function (lof) β-spectrin allele, em6, and an allele that creates a truncated β-Spectrin protein, em21(Dubreuil et al., 2000), also have similar phenotypes (Fig. 1I). In addition, mutant p-element embryos show reduced levels ofβ-Spectrin protein (see Fig. S1 in the supplementary material). Taken together, these data indicate that the p-elements that cause axonal defects disrupt β-spectrin gene function.

To better characterize the embryonic midline axon guidance defects associated with mutations in β-spectrin, we stained wild-type and mutant embryos with two neuronal markers, BP102 and Fas2/1D4. In wild-type embryos (or β-spectrin heterozygous embryos) stained with BP102,the CNS has a uniform ladder-like axon scaffold, containing longitudinal axon bundles running along either side of the midline that are connected to each other by distinct anterior and posterior commissures in each segment(Fig. 1B). By contrast,β -spectrin mutant embryos have thicker anterior and posterior commissures that are often fused (Fig. 1C), phenotypes that are indicative of loss of midline repulsion. Because mutations that disrupt midline glial cell development and migration often give rise to phenotypes similar to those seen inβ -spectrin embryos, we used antibodies against Wrapper(Noordermeer et al., 1998),and confirmed that the midline glia are present in the appropriate number and positions (Fig. 1L). The midline glia in β-spectrin mutants also express Slit, suggesting they produce appropriate repulsive guidance cues (data not shown).

Often, when midline repulsion is compromised, the most medial Fas2-positive longitudinal fascicle inappropriately crosses the midline and continues to extend on the opposite side (Kidd et al.,1998). Therefore, we investigated whether β-spectrinmutants had inappropriate Fas2-positive bundles crossing the midline. Similar to wild type, β-spectrin heterozygous embryos have no inappropriate crossovers (Fig. 1E,I). By contrast, β-spectrin hemizygous mutant embryos display ectopic Fas2-positive neurons crossing the midline(Fig. 1F,I). em6, em21and the p-element allele G0108 show the strongest crossing defects,whereas the presumptive hypomorphic p-element alleles (G0198 and G0074) have less severe defects(Fig. 1I). Additionally, when compared with wild type, β-spectrin mutant embryos show a reduced distance between opposing Fas2-positive medial longitudinal bundles(Fig. 1E,F), and the most lateral Fas2-positive bundles of axons appear to stall and fail to extend longitudinally (Fig. 1F,arrowheads). In C. elegans, mutations in β-spectrin(unc-70/bgs-1) also appear to affect neuronal function and axon outgrowth (Hammarlund et al.,2000; Moorthy et al.,2000), which suggests that Drosophila β-Spectrin may be playing a conserved role during the outgrowth of certain classes of axon.

To investigate where β-Spectrin functions during midline axon guidance, we examined β-Spectrin protein expression in wild-type andβ -spectrin mutant embryos using an antibody that specifically recognizes full-length β-Spectrin(Byers et al., 1989; Dubreuil et al., 2000). Wild-type β-Spectrin localizes weakly to the axon scaffold and is targeted to the plasma membrane surrounding every cell within the midline neuropil (Fig. 1J). em6 embryos have greatly reduced levels of β-Spectrin protein(Fig. 1K), yet we cannot rule out the possibility that trace levels may be maternally loaded. Strong and weak loss-of-function p-element alleles exhibit reduced levels and mislocalization (including lack of complete uniform cell body plasma membrane staining) of β-Spectrin protein (see Fig. S1 in the supplementary material).

The ubiquitous expression pattern of β-Spectrin did not give any insight into where β-Spectrin is required for its midline guidance function. Therefore, to investigate this, we tested two candidate cell types -neurons and midline glia - by expressing wild-type β-Spectrin protein in both cell types using the Gal4/UAS system in β-spectrin mutants. We found that the mutant phenotypes can be rescued by expressing wild-type UAS-β-Spectrin in all postmitotic neurons using elavGAL4(Fig. 1D,G). By contrast,expressing β-Spectrin in midline glia under the control of simGAL4 does not rescue the β-spectrin phenotype(Fig. 1H). Importantly,overexpression of β-Spectrin throughout the CNS in a wild-type background using elavGAL4 does not result in any obvious defects (data not shown). These data demonstrate that β-Spectrin is required in neurons but not in other cell types at the midline, and support the idea thatβ-Spectrin contributes to axon repulsion.

Canonically, α- and β-Spectrin function together as a heterotetramer that ubiquitously associates with the plasma membrane(Bennett and Baines, 2001; De Matteis and Morrow, 1998). Therefore, we investigated whether α-Spectrin, like β-Spectrin, has a role during midline axon guidance. In contrast to what is observed inβ -spectrin mutant embryos, α-spectrin mutants display only very mild defects in axon guidance. Specifically, these errors are seen just in certain hetero- and homoallelic combinations(Fig. 2A,C,F). When both copies of α-spectrin are removed in a hypomorphic hemizygousβ -spectrin background, we see an increase in the number of Fas2-positive axon bundles ectopically crossing the midline(Fig. 2D-F). This genetic interaction supports the idea that α-spectrin modestly contributes to axon guidance. However, β-Spectrin appears to be playing a much larger role during this developmental process. Hulsmeier et al. independently identified additional α-spectrin alleles that do not display guidance defects(Hülsmeier et al., 2007). These data further support the idea that α-spectrin plays only a minor role during midline guidance.

To investigate the relationship between α- and β-Spectrin, we examined their expression patterns in wild-type and single mutant backgrounds. Wild-type α- and β-Spectrin colocalize at the plasma membrane around every cell in the CNS and are weakly expressed on axon tracks(Fig. 3A-C). As noted above,β -spectrin (em6) mutant embryos exhibit severely reduced levels of β-Spectrin protein(Fig. 3D). Similarly, when image parameters are kept constant, α-Spectrin protein levels are also greatly reduced in β-spectrin mutants. However, the Photo Multiplier Tube (PMT) gain can be increased to detect low levels ofα-Spectrin at the plasma membrane surrounding the cell bodies of the CNS, although the protein looks slightly more diffuse than in wild type(Fig. 3E, starred arrowheads). The fact that levels of α-Spectrin are dramatically reduced inβ -spectrin mutant embryos suggests that localization and/or maintenance of α-Spectrin depends on β-Spectrin. However, the fact that we still see some α-Spectrin would suggest that there may be a minimal supply of maternally deposited β-Spectrin still present (although barely detectable by our antibody) in the β-spectrin mutant embryos that is able to localize α-Spectrin. We also cannot rule out the alternative possibility that there is a small pool of α-Spectrin that can localize independently of β-Spectrin. When we stain embryos carrying the hypomorphic allele (G0108) with anti-β-Spectrin and anti-α-Spectrin, we see reduced levels and mislocalization of both proteins. Intriguingly, α-Spectrin still colocalizes withβ-Spectrin (see Fig. S1 in the supplementary material) even though the overall expression pattern is highly disrupted. This is consistent with the hypothesis that β-Spectrin contributes to the localization and/or maintenance of α-Spectrin.

As expected, α-spectrin mutants have reduced levels ofα-Spectrin protein (Fig. 3H). The residual amount of α-Spectrin protein present in the embryo appears to still localize properly to the cell body plasma membrane, but not to axons (Fig. 3H, arrows), and is likely to be maternally contributed. In marked contrast to the clear dependence of α-Spectrin expression and/or stability on wild-type β-Spectrin, α-spectrinmutants exhibit a noticeable redistribution of β-Spectrin to the axon scaffold (compare Fig. 3A with 3G). Thus, it appears that α-Spectrin is normally required to keep axonal levels of β-Spectrin low.

Studies of βIV-Spectrin and Ankyrin-G mutant mice suggest that these two proteins may mutually stabilize each other, as well as CAMs and VGSCs, at axon initial segments and nodes of Ranvier(Komada and Soriano, 2002). Therefore, we were interested in testing whether there was a similar relationship between Ankyrin and β-Spectrin in Drosophila. Currently, dank1 (Drosophila ankyrin-1; ankyrin -FlyBase) mutants are not available but their interaction can be studied by using a mutant β-Spectrin transgene in which Ankyrin-binding activity has been removed (Das et al.,2006). Specifically, this βSpecΔank transgene cannot rescue the localization defects of Dank1::EGFP in copper cells ofβ -spectrin mutants, suggesting that Ank1-binding is indeed disrupted. Surprisingly, when expressed ubiquitously throughout the embryo,βSpecΔank rescues the axon guidance defects as efficiently as does wild-type β-Spectrin (Fig. 4C,D,F). Because we know that β-Spectrin is required in neurons (Fig. 1), we attribute this rescue to its neuronal function. These data suggest that the role ofβ-Spectrin during midline axon guidance in Drosophila is independent of Dank1.

Interestingly, Ankyrin-independent membrane binding sites forβ-Spectrin were identified in binding studies with NaOH-stripped membranes from rat brain (Davis and Bennett, 1994), and a reporter consisting of the PH domain of vertebrate βI-Spectrin fused to GFP binds IP3 and mediates targeting to the plasma membrane in COS cells (Wang et al., 1996). Therefore, to address whether the Ankyrin-binding or PH domains are required for precise localization of β-Spectrin in Drosophila neurons, we used the Myc epitope tag to monitor transgene expression. As expected, ubiquitously expressed full-length β-Spectrin localizes to axons and the plasma membrane of all cells in the neuropil(Fig. 4H,K). Similarly,βSpecΔank localizes with the same pattern as the full-length protein (Fig. 4I,L). By contrast, βSpecΔPH does not localize to the plasma membrane of cell bodies and the overall level of expression is reduced when compared with the other transgenes (Fig. 4J,M). However, we still observe some axonal staining(Fig. 4J). Despite disrupted cell body plasma membrane staining, βSpecΔPH completely rescues the midline guidance defects observed in β-spectrin mutant embryos(Fig. 4E), suggesting that the PH domain is not required for β-Spectrin to carry out its role in guidance. Together, these data suggest that the PH domain, but not the Ankyrin-binding domain, is required for proper membrane localization in the cell body. However, cell body plasma membrane localization does not appear to be required for function, as βSpecΔPH can still rescue the guidance phenotype despite partially defective distribution. This is consistent with the idea that a sub-population of axonally localized β-Spectrin is important during growth cone migration, and this localization is independent of the PH and Dank1-binding domains. However, we do acknowledge that these gain-of-function experiments may not reflect the true localization of endogenous proteins with similar deletions. For example, in our experiments,overexpressed full-length β-Spectrin appears more concentrated on axons than does endogenous protein (compare Fig. 1J with Fig. 4H).

As has been previously shown, Fas2-positive axons ectopically cross the midline in β-spectrin mutant embryos. Given the similarity between β-spectrin and robo loss-of-function phenotypes, we speculated that β-spectrin might be influencing Slit-Robo-mediated repulsion. Therefore, we investigated whether mutations inβ -spectrin could enhance the guidance defects associated with heterozygous mutations in slit and robo. Observing a dose-dependent genetic interaction between a set of genes suggests that they may be involved in the same pathway. Indeed, mutations inβ -spectrin demonstrate a genetic interaction with slitand robo. Specifically, β-spectrin hemizygous em6 mutants, or slit, robo trans-heterozygous mutants,contain few ectopic Fas2-positive bundles crossing the midline(Fig. 5A,B,G). However, in combination, β-spectrin hemizygous mutations dramatically enhance the slit, robo trans-heterozygous phenotype(Fig. 5C,G), suggesting that these genes are involved in the same process, namely midline repulsion. Importantly, hypomorphic β-spectrin alleles (G0198 and G0074) also robustly enhance the slit, robotrans-heterozygous phenotype (Fig. 5E-G). These dose-dependent interactions suggest thatβ -spectrin may be intimately involved in the Slit-Robo pathway. Furthermore, β-spectrin mutants seem to specifically modify the Slit-Robo pathway, as genetic interactions are not detected with either the Netrin/Frazzled or the Semaphorin/Plexin pathways (see Fig. S2 in the supplementary material).

Similar to the rescue of β-spectrin mutants observed by expressing wild-type β-Spectrin protein in all neurons, we also see rescue of the slit, robo enhancement by expressing wild-typeβ-Spectrin with elavGal4(Fig. 5D,G). Given these genetic interactions, we hypothesized that β-Spectrin may regulate the distribution of Robo in Drosophila midline CNS axons. However, we did not observe obvious changes in the levels or localization of Robo protein whenβ-spectrin mutant embryos were compared with wild type(Fig. 6). Nevertheless,β-Spectrin may affect another component of the Robo pathway downstream of the receptor.

In addition to defects in Fas2-positive neurons, we also seeβ -spectrin mutant phenotypes and genetic interactions with the Slit-Robo pathway within a small subpopulation of ipsilateral neurons, the apterous (Ap) neurons. In wild-type embryos, the Ap neurons approach the midline but never cross (Fig. 7A). In both slit/+ heterozygous andβ -spectrin (em6) hemizygous mutant embryos, we observe mild midline errors (`thick crosses') when Ap neurons are labeled with Tau-Myc-GFP (Fig. 7B,C,H). Both null and hypomorphic mutations in β-spectrin enhance the slit/+ heterozygous phenotype(Fig. 7D,H). In contrast to pan-neural expression, which completely rescues the mutantβ -spectrin guidance defects, the Ap neuron defect cannot be rescued by expressing full-length β-Spectrin in the Ap neurons(Fig. 7E,H). These data suggest a non-autonomous function for Spectrin for some classes of neuron and support the idea that the overall defective patterning of the entire CNS contributes to defects seen in the Ap neurons. robo mutants also display ectopic Ap neurons crossing the midline (Fig. 7F); however, this defect can be rescued by expressing full-length Robo in these neurons, demonstrating that Robo functions cell autonomously(Fig. 7G).

In the course of examining the strong midline crossing phenotypes that result from genetic interactions between β-spectrin and slit and robo, we were struck by the absence of defects in the early guidance of the pCC and dMP2 pioneer axons. Initially inβ -spectrin mutant embryos, pCC and dMP2 make appropriate ipsilateral projections, and only later are crossing defects observed (data not shown). This is true even in the context of genetic interactions betweenβ -spectrin and slit and robo, where severe midline crossing defects are observed later in development. This raised the possibilities either that guidance of the pioneer axons does not requireβ -spectrin function, whereas later navigating axons do, or alternatively that β-spectrin function is required for the maintenance of properly established connections. As it is no longer possible to resolve individual neurons at the time that midline crossing defects are observed with Fas2 antibodies, we analysed the behavior of the Ap neurons more closely to distinguish between guidance and maintenance functions.

Similar to what we have observed for the early behavior of the pCC and dMP2 axons, axons of the Ap neurons initially make the correct steering decisions in β-spectrin mutant embryos. Specifically, projection toward the midline is normal in early stage 14 embryos and axons make correct anterior ipsilateral turns at late stage 14 to early stage 15(Fig. 8G,H), similar to wild type (Fig. 8A-C). However midline-crossing defects are observed by late stage 16(Fig. 8I). By contrast, robo mutants exhibit Ap neuron crossing defects during their initial extension toward the midline and throughout development(Fig. 8D-F). In addition,although β-spectrin mutants that are also heterozygous for slit show considerably stronger guidance and crossing defects, these defects only arise at later stages of CNS development(Fig. 8K,L). These observations suggest (1) that Slit-Robo repulsion is required not only for initial axon guidance, but also to maintain connections on the correct slide of the midline, and (2) that β-spectrin contributes to this later maintenance function.

In this study, we demonstrate that Drosophila β-Spectrin is required exclusively in neurons for normal repulsive midline axon guidance. Limiting β-spectrin shows a dose-dependent synergistic interaction with trans-heterozygous mutations in slit and robo. This genetic interaction suggests that β-Spectrin may be specifically contributing to Slit-Robo repulsion during growth cone migration. Our work also demonstrates that neither the Ankyrin-binding nor the Pleckstrin Homology (PH) domains of β-Spectrin are required for its role in repulsive guidance. Intriguingly, axonal localization is not disrupted whenβSpecΔPH or βSpecΔAnk mutant proteins are overexpressed. These overexpression experiments suggest that β-Spectrin may localize to axons independently of the Ankyrin-binding and PH domains, and that axonalβ-Spectrin is sufficient for growth cone migration. However, it is important to note that because these observations are based on overexpression studies, they may not completely reflect the localization of endogenous proteins with similar deletions. Finally, analysis of progressive developmental stages in β-Spectrin mutants suggests that, for at least some neurons, β-Spectrin is not required for the initial pathfinding of axons, but rather appears to be important for maintaining appropriate connectivity. Together, our findings support a role of the Spectrin network in contributing to the fidelity of axon repulsion at the Drosophila midline.

The midline guidance defects we observe in embryos stained with BP102 and Fas2 antibodies suggest that β-Spectrin normally contributes to axonal migration and more specifically to axon repulsion. Considering thatβ-Spectrin has been shown to modulate the behavior of interacting membrane proteins within sub-membrane microdomains, and given that we observe specific dose-dependent genetic interactions with the Slit-Robo pathway, our data support the idea that the Spectrin cytoskeleton modulates the behavior of molecules that contribute to Robo repulsion. Many important signaling molecules must be coordinated downstream of guidance receptors so that navigating growth cones make appropriate decisions. For example, Dock, Pak and Rac contribute to midline repulsion by forming a complex with the Robo receptor upon Slit stimulation (Fan et al., 2003). Similarly, Drosophila Ena, a member of a protein family implicated in actin cytoskeleton regulation, functions cooperatively with Robo at a level downstream of the receptor(Bashaw et al., 2000). Additionally, the microtubule-associated protein CLASP (Chb - FlyBase) and the Abelson tyrosine kinase are required to induce restricted cytoskeletal events at the leading edge of growth cones (Lee et al., 2004). Along these lines, β-Spectrin may affect the ability of comparable proteins to signal effectively in the Slit-Robo pathway. Clearly, the Spectrin network cannot account for all Robo function, as many ipsilateral CNS neurons are still guided properly in β-spectrinmutants. Mutations in β-spectrin also enhance a robo1null loss-of-function mutation (data not shown), suggesting that the Spectrin network probably plays an additional role in repulsion outside of Robo1 signaling. It is important to note that our data do not rule out the alternative possibility that β-spectrin mutations may have general effects on growth cone migration. If so, then the specific interactions seen with slit and robo might reflect the fact that the Slit-Robo pathway is more susceptible to subtle perturbations than are the other guidance pathways tested.

Our analysis of the effect of β-Spectrin mutations on the guidance of Ap neurons suggests that β-Spectrin is required not for the initial pathfinding of these axons, but rather for maintaining correctly established connections. Moreover, our genetic analysis suggests that in this context the Slit and Robo system also contributes to this maintenance function. Our observations support the idea that some level of Slit/Robo repulsion is required continuously to keep ipsilateral axons on their own side of the midline. A more dramatic example of pathway maintenance has been described in C. elegans, where mutations in the Zig genes lead to a `flipping' of axon pathways over the midline(Aurelio et al., 2003; Aurelio et al., 2002). Additionally, in Drosophila, it has been shown that β-Spectrin is essential for synapse stabilization(Pielage et al., 2005). It will be interesting in the future to determine whether the maintenance function of slit, robo and β-Spectrin represents a repulsive mechanism distinct from the mechanism operating during pathway formation.

Our work demonstrates that in the absence of β-Spectrin the stability of α-Spectrin is decreased. Thus, it appears that hetero-tetramer formation is required to maintain proper levels of α-Spectrin in the nervous system. By contrast, cell body plasma membrane localization ofβ-Spectrin is unaffected in α-spectrin mutants. Consistent with this result, it has been shown that β-Spectrin accumulates independently of α-Spectrin in Drosophila larvae(Dubreuil and Yu, 1994). This suggests that β-Spectrin recruitment to and stability at the cell body plasma membrane are independent of α-Spectrin, and supports the idea of an α-Spectrin-independent role for β-Spectrin in neurons (see below).

If α- and β-Spectrin function together as a hetero-tetramer, why then do we observe only mild axon guidance defects inα -spectrin single mutants? One explanation is that perhapsα-Spectrin has a higher maternal component that is able to compensate for a lack of zygotic expression. Indeed, α-spectrin mutants survive to a later developmental stage than do β-spectrinmutants. However, α-spectrin mutants also exhibit increased levels of axonally localized β-Spectrin (see below). Therefore, an alternative hypothesis could be that the preferential distribution ofβ-Spectrin to axons somehow compensates for a reduction ofα-Spectrin, allowing neurons to make precise steering decisions. Lastly,and perhaps most interesting, β-Spectrin may function independently ofα-Spectrin in neurons. This idea seems plausible given thatβ-Spectrin remains properly localized to the cell body plasma membrane inα -spectrin mutants.

As noted above, α-spectrin mutant embryos show an increase in the levels of axonal β-Spectrin, suggesting that α-Spectrin regulates the accumulation of β-Spectrin in axons. What does this shift in β-Spectrin distribution mean for the neuron? What are the signals that target β-Spectrin to axons and why does the localization change inα -spectrin mutants? One possibility is that the SH3 domain ofα-Spectrin targets β-Spectrin to the cell body plasma membrane via a direct interaction with another protein. Another possibility is that, in the absence of α-Spectrin, β-Spectrin binds other axonally localized proteins to a greater degree, thus shifting the overall distribution.

It was originally thought that Ankyrin-G is assembled upstream ofβIV-Spectrin at axon initial segments(Jenkins and Bennett, 2001);however, later reports suggested that both Ankyrin-G and βIV-Spectrin are required for the localization and stability of one another, as well as for the stability of VGSCs at axon initial segments and nodes of Ranvier(Komada and Soriano, 2002). Yet, in neonatal cardiomyocytes, Ankyrin-B is required for the proper distribution and levels of βII-Spectrin, and an Ankyrin-B protein lacking the Spectrin-binding domain still localizes properly. These data suggest that Ankyrin-B localization is independent of βII-Spectrin. We sought to test the functional importance of the protein domains of β-Spectrin during axon guidance, and the order of assembly of β-Spectrin and Ankyrin in Drosophila neurons.

We observe that a mutant form of β-Spectrin missing the Drosophila Ankyrin-binding domain (βSpecΔank) remains properly localized to cell bodies and axons. Importantly, a recent study has shown that the same βSpecΔank protein used in this study correctly accumulates at the plasma membrane of copper cells yet fails to accurately target an Ank-GFP fusion protein in epithelial cells(Das et al., 2006), suggesting that Ankyrin localization depends on β-Spectrin but not vice versa. In addition, βSpecΔank rescues the guidance errors seen inβ -spectrin mutants. Taken together, our data suggest that Ankyrin-binding is not essential for β-Spectrin localization and guidance function in neurons, and that Drosophila β-Spectrin may be assembled upstream of Ankyrin.

It is important to note that Drosophila has another Ankyrin gene, dank2 (ank2 - FlyBase), which is expressed specifically in neurons (Hortsch et al.,2002). Therefore, perhaps accurate β-Spectrin targeting occurs via a direct interaction with Dank2. Keeping in mind thatβSpecΔank is missing the conserved Ankyrin1-binding site, this logic would imply that β-Spectrin associates with Dank2 using a different binding site than that used for Dank1. There does appear to be an intimate relationship between Dank2 and β-Spectrin, as Dank2 is mislocalized and levels are reduced in β-spectrin mutants (D.S.G. and G.J.B.,unpublished). Future experiments will help to establish whether Dank2 can bindβ-Spectrin at a location different than that used by Dank1. If the mutantβSpecΔank does indeed bind to Dank2, it will be interesting to determine the functional significance of this interaction.

We also demonstrated that, although the PH domain of Drosophilaβ-Spectrin is not necessary for axonal localization and/or guidance function, it is required for appropriate localization to the cell body plasma membrane. From our experiments, it appears that axonally localizedβ-Spectrin is important for making accurate guidance decisions. In the future, it will be important to determine the region of β-Spectrin that is essential for axonal targeting and whether this domain is important for growth cone migration.

At nodes of Ranvier and axon initial segments, there is an intimate relationship between the Spectrin network, Ankyrin and CAMs(Lambert et al., 1997). For example, in Purkinje and granule cells, Ankyrin-G localization precedes that of neural CAMs, and is required for proper CAM cluster assembly at axon initial segments (Jenkins and Bennett,2001; Zhou et al.,1998). Other studies of CAMs have revealed that they can commonly act as contactmediated attractive and repulsive signals for growing axons(Doherty and Walsh, 1994; Tanaka and Sabry, 1995). In Drosophila, pre-synaptically targeted β-spectrin RNAi disrupts the stability and organization of CAMs such as Fas2 and Neuroglian(Pielage et al., 2005), and neuroglian mutants exhibit defects in motoneuron pathfinding(Hall and Bieber, 1997). Furthermore, neurotactin mutant embryos exhibit midline axon guidance errors that appear to be similar to those observed in β-spectrinmutants, including ectopic midline crossing and longitudinal breaks(Speicher et al., 1998). Future studies should investigate whether there are functional links between CAMs and the Spectrin network in the context of midline axon guidance.

We thank Alex Kolodkin and the Bloomington Stock Center for fly stocks. We are grateful to the Bashaw lab for constructive comments and technical assistance/support. This work was supported by NIH GM 49301 to R.R.D. and NIH Grant NS046333 to G.J.B.