The F-actin-microtubule crosslinker Shot is a platform for Krasavietz-mediated translational regulation of midline axon repulsion

The developmentally stereotyped migrations of neuronal growth cones establish an initial network of neuronal connections that is crucial for proper nervous system function. Chemical gradients or local cues presented on cellular landmarks guide these migrations by controlling dynamic rearrangements of the growth cone cytoskeleton(Dent and Gertler, 2003; Dickson, 2002). As growth cones encounter these cues, they adapt their responses. When reading a chemical gradient, growth cones continuously reset their threshold sensitivity to the cue so that higher concentrations elicit cytoskeletal responses required for motility and guidance (Song and Poo, 2001). Locally presented cues may also qualitatively change growth cone responsiveness, enabling them to proceed to the next cellular target in the pathway (Song and Poo, 2001).

Growth cone adaptation may require new local protein synthesis(Dickson, 2002). Axonal or dendritic transport may not be fast enough to supply proteins to growth cones at a distance from the cell body. Axotomy experiments with protein synthesis inhibitors indicate that new protein synthesis in axons or their growth cones is required for turning in vitro in response to extracellular gradients of guidance cues (Campbell and Holt,2001; Ming et al.,2002). Local protein synthesis is also required for growth cones to change their responsiveness to specific local cues. In the spinal cord of vertebrates, translation of ephA2 receptor mRNA is locally regulated,occurring in commissural growth cones only after they cross the midline(Brittis et al., 2002). Translational regulation in axons and their growth cones may involve rapamycin- or MAP kinase (MAPK)-sensitive pathways(Campbell and Holt, 2001; Campbell and Holt, 2003; Ming et al., 2002), and also cytoplasmic polyadenylation element (CPE)-dependent mechanisms(Brittis et al., 2002). However, the mechanism connecting receptor signaling to translation is largely unknown.

In the Drosophila embryo, the midline repellent Slit prevents longitudinally directed CNS axons from crossing the midline(Kidd et al., 1999) and directs them into particular longitudinal pathways through its Robo, Robo2 and Robo3 receptors on the CNS growth cones(Kidd et al., 1999; Simpson et al., 2000). As we show here, the spectraplakin Short stop (Shot; also known as Kakapo) is also required for midline axon repulsion. Mutations in shot lead to ectopic midline crossing of Fas II-positive axons and dominantly enhance the slit or robo heterozygous loss-of-function phenotypes,suggesting that shot may function in the same guidance process as slit and robo.

Shot mediates direct interactions between F-actin and microtubules (MTs)required for initial sensory axon extension and motor axon extension to target muscles (Lee and Kolodziej,2002b). We provide evidence that the role of Shot in midline guidance involves translational regulation. Shot physically interacts with Krasavietz (Kra; also known as Exba), an evolutionarily conserved putative translation factor, and this interaction is required for Shot activity to support midline axon repulsion. Kra contains a C-terminal W2 domain found in the translation initiation factors eIF5, eIF2Bϵ, DAP-5 and eIF4G. The W2 domains in these translation initiation factors have been shown to mediate protein-protein interactions among translation factors that are required for preinitiation complex assembly (Preiss and Hentze, 2003). As suggested by the presence of W2 domain, Kra binds to eIF2β and 40S ribosomal subunits. We also show that Kra inhibits translation in vitro. Finally, we show that in kra or eIF2β mutant embryos some Robo-expressing axons lose the ability to respond to the midline repellent Slit. Taken together, our data suggest that Shot functions as a platform for translational control of midline axon guidance. Through this proposed Shot-Kra-eIF2β circuit, the translation of mRNAs encoding proteins essential for axon guidance at the midline may be closely tied to the pace of cytoskeletal assembly.

All full-length cDNA clones for kra, eIF5 and eIF2βwere obtained from the Drosophila Genomics Resource Center (GenBank accession numbers AA440023, AW941924 and BI23305, respectively). For glutathione S-transferase (GST) pull-down assays, the full-length coding sequences of kra and eIF5 were amplified by polymerase chain reaction (PCR) and subcloned into pGEX6P1 (Pharmacia). For in vitro translation, the full-length coding regions of eIF2β were subcloned into pCDNA3.1 (Invitrogen). The C-terminal domain of the Shot long isoforms, C-Shot L [amino acids (AA) 4689 to 5201; GenBank accession number AAF24343], and its deletions were amplified by PCR and subcloned into pSP64 Poly(A) (Promega). For the kra rescue experiments, the coding sequence of kra was tagged with hemagglutinin (HA) and subcloned into pUASTNEXT (Lee and Kolodziej,2002a), a derivative of the pUAST vector(Brand and Perrimon, 1993), to give UAS-HA-kra^WT. The wild-type kra cDNA was manipulated using the QuikChange Multi kit (Stratagene)to introduce the 12A and 7A mutations (Fig. 1B). The insert of UAS-HA-kra^WT was replaced with kra^12A and kra^7A cDNAs to give UAS-HA-kra^12A and UAS-HA-kra^7A, respectively. To overexpress eIF2β in the fly embryo, the entire coding sequence was tagged with green fluorescent protein (GFP) and introduced into pUASTNEXT.

Two P-element insertions in the kra locus, l(3)j9b6(Bloomington Stock Center, Bloomington, IN) and EP(3)0428 (Szeged Stock Center, Hungary), were obtained and mobilized by standard methods to generate kra¹ and kra², respectively. For the kra rescue experiments, transgenic flies expressing UAS-HA-kra^WT, UAS-HA-kra^7Aand UAS-HA-kra^12A cDNAs were obtained as previously described (Robertson et al.,1988). These transgenes were expressed in the kra¹/kra² mutant embryos under the control of da-GAL4 (Wodarz et al.,1995), C155-GAL4 (Lin and Goodman, 1994), repo-GAL4(Sepp et al., 2001) or slit-GAL4 (Scholz et al.,1997). shot rescue experiments were performed as previously described (Lee and Kolodziej,2002b). slit² and robo²were obtained from the Bloomington Drosophila Stock Center.

S2 cells were grown at 29°C in Schneider's medium (Invitrogen)supplemented with 10% heat-inactivated (30 minutes, 55°C) fetal calf serum. S2 cells were transfected using Cellfectin (Invitrogen) according to the manufacturer's instructions.

Double-stranded RNA interference (dsRNAi) was performed in six-well tissue culture plates containing 2×10⁶ S2 cells for 3 days as previously described (Clemens et al.,2000). Briefly, DNA templates containing T7 promoter sequences at their 5′ and 3′ ends were amplified by PCR and transcribed with the Megascript T7 transcription kit (Ambion) to generate dsRNAs. Primers contained T7 promoter sequences upstream of the following: kra sense primer, 5′-TTGGTCCACCATCATGTCATT-3′, kra antisense primer, 5′-ACTGAAGCCATTCGACAAAC-3′; gfp sense primer,5′-ACGTAAACGGCCACAAGTTC-3′, gfp antisense primer,5′-GTCCTCCTTGAAGTCGATGC-3′. For the assay, dsRNAs were used at a final concentration of 37 nM.

In order to raise antibodies against the full-length Kra and eIF2βproteins, GST-Kra and MBP-eIF2β were expressed in Escherichia coli BL21 (Stratagene) and purified with glutathione-Sepharose 4B(Amersham Pharmacia) and amylose resin (NEB), respectively. GST-Kra was digested with PreScission protease (Amersham Pharmacia). Protein samples were subjected to SDS-polyacrylamide gel electrophoresis (SDS-PAGE), and the bands representing 50 kDa Kra and 82 kDa MBP-eIF2β were excised for the immunization of guinea pigs and rats, respectively.

Monoclonal antibodies (mAbs) against Fasciclin II (1D4), Repo (8D12),Wrapper (10D3), Robo (13C9) and Slit (C555.6D) were purchased from the Developmental Studies Hybridoma Bank (DSHB). Additional antibodies used in this study were rabbit anti-eIF4E(Nakamura et al., 2004), 3F10 rat monoclonal anti-HA (Roche), rabbit anti-β-galactosidase (Cappel),rabbit anti-GFP (Abcam), rabbit anti-horseradish peroxidase (HRP) (MP Biochemicals) and goat anti-L28 (Santa Cruz).

Whole-mount staining of embryos was performed as previously described(Lee et al., 2000; Spencer et al., 1998).

For GST pull-down assays, GST fusion proteins of Kra and eIF5 were produced in E. coli and purified using glutathione-Sepharose 4B (Amersham Pharmacia). eIF2β and the C-terminal regions of the Shot long isoforms were synthesized using an in vitro transcription/translation kit (Promega) in the presence of [³⁵S]methionine. The binding of GST-Kra to the C-terminal regions of Shot was performed in 20 mM Tris-HCl (pH 7.5), 1 mM CaCl₂, 1% NP-40, 150 mM NaCl, 5 mM DTT and 10% glycerol, or in 20 mM Tris-HCl (pH 7.5), 1 mM EGTA, 0.1% NP-40, 150 mM NaCl, 5 mM DTT and 10%glycerol. The additional binding experiment was performed in a binding buffer containing 20 mM Tris-HCl (pH 8.0), 200 mM NaCl, 1 mM EDTA and 0.5% NP-40.

For immunoprecipitation, fly embryos or S2 cells coexpressing HA-Kra with GFP-eIF2β, Shot L(A)-GFP or C-Shot L-GFP were homogenized in immunoprecipitation buffer [50 mM Tris-HCl (pH 8.0), 1% NP-40, 150 mM NaCl, 2 mM Na₃VO₄, 10 mM NaF, 10% glycerol, protease inhibitors], and then centrifuged at 12,000 g for 25 minutes at 4°C. Supernatants were precleared by incubation with protein G-Sepharose (Pierce) for 1 hour at 4°C. The samples were incubated with anti-GFP or anti-HA for 4 hours at 4°C and then incubated with protein G-Sepharose for 2 hours at 4°C. Beads were washed three times with immunoprecipitation buffer and boiled in SDS sample buffer. The eluates were subjected to western blotting.

For in vitro translation, rabbit reticulocyte lysate (Promega) was first preincubated with 2.4 μM bovine serum albumin (BSA), GST, GST-Kra, Kra or Kra-7A. Each translation reaction contained 50% rabbit reticulocyte lysate,1×TNT reaction buffer (Promega), 40 μM AA mixture without methionine(Promega), 10 μCi [³⁵S]methionine (Amersham Pharmacia), 20 U of RNase inhibitor (Promega) and 70 nM luciferase mRNA. After incubating the reaction for 90 minutes at 30°C, 20% of each reaction was removed and analyzed by SDS-PAGE and autoradiography to monitor the translation output.

S2 cell extracts were prepared in lysis buffer [50 mM Tris-HCl (pH 7.5),250 mM NaCl, 50 mM MgCl₂, 100 μg/ml heparin, 1 mM DTT and 0.5 mg/ml cycloheximide]. Cell lysates were layered on 7-47% (w/v) linear sucrose gradients in the lysis buffer, and centrifuged at 270,000 gfor 3 hours at 4°C in a Beckman SW41 rotor. Fractions were collected from the top of the gradient and analyzed for absorbance at 260 nm to locate the ribosomal subunits (40S, 60S and 80S) and polysomes. Portions of fractions were further analyzed by SDS-PAGE and western blotting.

Through alternative splicing and multiple promoter usage, shotencodes multiple, modularly assembled protein isoforms. Neuronal expression of either the Shot L(A) or Shot L(B) isoform in shot mutant embryos rescues defects in sensory and motor axon extension to target muscles(Lee and Kolodziej, 2002b). These isoforms contain both an N-terminal F-actin binding domain (ABD) and a C-terminal domain (C-Shot L) with two putative Ca²⁺-binding(EF-hand) and microtubule-binding (GAS2) motifs(Fig. 1A), all of which are essential for axon extension (Lee and Kolodziej, 2002b). To investigate the function of the C-terminal motifs, we performed a yeast two-hybrid screen with a Drosophilaembryonic cDNA library using C-Shot L (AAF24343, residues 4688 to 5201) as a bait. Two of the interacting clones encoded the C-terminal domain (residues 315 to 422) of a novel protein, Krasavietz (Kra), which has been annotated in FlyBase as CG2922. The predicted Kra protein is 51-52% identical to its Xenopus (AAH41729), zebrafish (AAH58875), mouse (AAH05466) and human(hKra/BZAP45; BAA02795) homologs (Fig. 1B and data not shown). However, no Kra homologs have been identified in yeast or Caenorhabditis elegans.

To confirm Shot-Kra interaction, we coexpressed either Shot L(A)-GFP or C-Shot L-GFP with HA-Kra in S2 cells and performed coimmunoprecipitation experiments. We found that Shot L(A)-GFP specifically coimmunoprecipitated with HA-Kra (Fig. 1C). C-Shot L-GFP also formed complexes with HA-Kra with affinities comparable to that of Shot L(A)-GFP (Fig. 1C),suggesting that the C-terminal domain of Shot is largely responsible for its interaction with Kra.

Next, we performed GST pull-down assays to determine the Kra-binding motif in C-Shot L. GST-Kra efficiently bound to fragments of C-Shot L containing either the EF-hand and GAS2 motifs (EF-GAS2) or the EF-hand motifs only (EF),whereas the GAS2-containing fragment by itself showed Kra binding activity at lower levels (Fig. 1D). By contrast, ΔEF-GAS2, which removes both the EF-hand and the GAS2 motifs from C-Shot L, did not bind to GST-Kra(Fig. 1D). The importance of the EF-hand and GAS2 motifs for Shot-Kra interaction was further evaluated in S2 cells. We found that binding of Shot L(A)-GFP to HA-Kra was reduced to 5%by deletion of the EF-hand motifs and to 35% by deletion of the GAS2 motif(see Fig. S1 in the supplementary material). These observations indicate that the EF-hand motifs are the major Kra-binding sites, and that the GAS2 motif is also required for strong interactions. Interestingly, Shot L(A)-ΔEF-GFP bound endogenous α-tubulin at comparable levels to Shot L(A)-GFP (see Fig. S1 in the supplementary material), suggesting that Kra binding through the EF-hand motifs does not impair the ability of Shot to bind microtubules through the GAS2 motif.

The two-hybrid screen result indicates that the C-terminal domain of Kra contains the Shot binding site. A database search revealed that the region belongs to the W2 domain family found in translation initiation factors. Multiple alanine substitutions (12A and 7A in the AA-boxes 1 and 2,respectively) in the W2 domain of eIF5 disrupt its interactions with eIF2β and eIF3-NIP1 (Asano et al.,1999). To pinpoint the Shot binding site with regard to possible translation initiation factor binding sites, we made similar alanine substitutions in Kra (Fig. 1B). The Kra-12A mutant, as well as the wild-type Kra, efficiently bound C-Shot L in S2 cells, whereas the binding of C-Shot L with the Kra-7A mutant was minimal, appearing only at background levels(Fig. 1E). These data suggest that the evolutionarily conserved residues in the AA-box 2 motif of Kra are essential for Shot-Kra interaction.

We developed an antibody against full-length Kra, which detects a single 50-kDa band on western blots containing extracts of S2 cells(Fig. 2A). Levels of the recognized protein were significantly reduced by kra dsRNAi, but not by gfp dsRNAi, suggesting the specificity of our anti-Kra antibody(Fig. 2A). Immunostaining using anti-Kra revealed that Kra was highly expressed in early stage embryos (e.g. stage 3; Fig. 2B) because of a significant maternal contribution. It was detected in the CNS and epidermis by stage 11, and in the gut by stage 13 (Fig. 2C). High expression in the CNS was maintained until the end of embryogenesis (Fig. 2D).

We inspected the embryonic CNS by double labeling with anti-Kra and anti-Elav, a neuronal marker (Lin and Goodman, 1994), and found that Kra is expressed in most or all post-mitotic neurons (Fig. 2F-H). Kra localized primarily to the cytoplasm of those cells,but was not clearly detectable in CNS axons(Fig. 2H). Kra was also expressed in many CNS glial cells (Fig. 2I-K), as assessed by double labeling with anti-Kra and anti-Repo,a glial marker that visualizes CNS glial cells, except for the midline glia(Halter et al., 1995). To assess the expression of Kra in midline glial cells, we used a midline glial marker anti-Wrapper (Noordermeer et al.,1998). However, because of ubiquitous expression of Kra around the midline, we were not able to determine its presence in these cells. We next wished to examine the localization of Kra at single axon resolution. We therefore expressed UAS-HA-Kra^WT in a small subset of embryonic CNS neurons with the aid of a neuronal driver apterous (ap)-GAL4(O'Keefe et al., 1998). HA-Kra was strongly detected in the cell body and the axon of the Ap neurons(Fig. 2L), suggesting that Kra can be localized into axons.

Partial loss-of-function P-element alleles of kra affect learning and memory (Dubnau et al.,2003). To determine its role in neuronal development, we generated molecularly defined kra-null mutants. We first verified that the transposons of l(3)j9B6 and EP(3)0428 were inserted at -1636 and -1228, respectively, of the kra open reading frame (ORF), which is common to its multiple transcript variants(Fig. 3A). We then mobilized these P-elements, recovered 200 independent excision lines from each parental allele, and determined the breakpoints of 30 lethal alleles by PCR and DNA sequencing. In particular, we found that kra¹, derived from l(3)j9B6, deletes 2068 bp (-1729 to +338), whereas kra², derived from EP(3)0428, deletes 1404 bp(-1242 to +161). Both deletions remove most of the 5′-UTR and the translation start site, and extend into the ORF(Fig. 3A). Homozygotes and transheterozygotes for these deletion alleles died at the pupal stage. The anti-Kra antibody cleanly detected the Kra protein in wild-type, but not in kra¹/kra² mutant larvae or pupae(Fig. 3B and data not shown),suggesting that both kra¹ and kra² are null alleles for kra.

The kra gene is located 463 bp upstream of the putative gene CG1427, whose function is unknown. Reverse transcription (RT)-PCR analysis showed that levels of CG1427 mRNA were normal in kra¹/kra² third instar larvae(Fig. 3C), suggesting that kra¹ and kra² mutations do not affect the expression of CG1427.

We next investigated whether axon extension and guidance are defective in kra-null mutant embryos. No defects in sensory and motor extension were observed in kra¹/kra² embryos stained with mAb 22C10 or 1D4 (data not shown). We then examined axon phenotypes in the CNS. In wild-type embryos at early stage 13, mAb 1D4 labeled the pCC axon that pioneers the ipsilateral pCC pathway without crossing the midline(Fig. 4A). In kra¹/kra² embryos, the same axon aberrantly often crossed the midline (Fig. 4B). This early-stage axon pathway defect indicates that kra is required for accurate growth cone migration. This phenotype becomes more obvious in later stage embryos. In wild-type embryos at stage 16,mAb 1D4 labeled three longitudinal axon pathways on each side of the midline;axons in these pathways did not cross the midline(Fig. 4D). In kra¹/kra² embryos, axons from the innermost(pCC) pathway ectopically crossed the midline in 18% of CNS segments(Fig. 4E,L). Kra is ubiquitously expressed in glial cells and neurons of the embryonic CNS. To determine where Kra functions in midline axon guidance, we examined CNS axon development in kra¹/kra² embryos that express UAS-HA-kra^WT under a neuron-specific driver C155-GAL4. In these embryos, only 6% of CNS segments exhibited the midline crossing defect (Fig. 4F,L), suggesting that Kra functions, at least in part, in neurons to enable growth cones to avoid the midline. By contrast, glial-specific expression of UAS-HA-kra^WT using slit-GAL4 or repo-GAL4 did not improve the kramidline phenotype (19% with slit-GAL4, n=464; 18% with Repo-GAL4,n=512).

kra-null mutant embryos contain substantial amounts of maternally contributed Kra protein. To generate embryos that lack maternally and zygotically contributed Kra, we used the FRT/ovoD1 method(Chou et al., 1993) to generate females containing homozygous kra¹ germline clones. These females did not lay eggs, and dissected germlines were blocked in oogenesis(data not shown). Kra therefore appears to be required during oogenesis,precluding isolation of embryos lacking maternally and zygotically contributed Kra.

In shot³ mutant embryos, motor axons extend outward from the CNS, choosing the right pathways but then stalling short of their muscle targets (Lee et al.,2000). Sensory axons also extend appropriately during early parts of their trajectory but fail to advance(Lee et al., 2000). CNS axon phenotypes for shot have not been previously described. In the present study, we found that Fas II-positive CNS axons ectopically cross the midline in ∼16% of segments in shot³ embryos(Fig. 4C,G), suggesting the role of Shot in midline axon guidance. Neuronal expression of Shot L(A)-GFP rescued the midline crossing phenotype of shot³ embryos(Fig. 4L). We have shown previously that the F-actin-microtubule crosslinking activity of Shot is essential for axon extension (Lee and Kolodziej, 2002b), and therefore we investigated whether this would still be the case for midline axon guidance. The microtubule-binding domain mutant Shot L(A)-ΔGAS2-GFP rescued the CNS phenotype of shot³ embryos (Fig. 4H,L), although the same transgene did not rescue their axon extension phenotypes as was previously described(Lee and Kolodziej, 2002b). By contrast, the F-actin-binding mutant Shot L(C)-GFP(Lee and Kolodziej, 2002b) was unable to rescue the loss of shot function in the embryonic CNS(Fig. 4I,L). These results suggest that midline axon repulsion requires the F-actin-binding activity of Shot, but neither the microtubule binding nor the F-actin-microtubule crosslinking activity. Shot L(A)-ΔEF-GFP showed only partial activity to improve the midline crossing defect in shot³ mutant embryos (Fig. 4L), suggesting that Kra binding through the EF-hand motifs is also important for the role of Shot in midline axon repulsion.

To determine the physiological relevance of the biochemical interaction of Kra with Shot and phenotypic overlap between kra and shot at the midline, we investigated genetic interactions between kra and shot. The frequency of midline crossovers was significantly worse in shot³/+; kra¹/kra² and shot³/shot³; kra²/+ mutant embryos than in either homozygous kra¹/kra² or shot³/shot³ mutant embryos(Fig. 4J-L). The midline crossing phenotype in shot³/shot³; kra¹/kra² double mutant embryos was slightly worse than that in shot³/shot³; kra²/+ mutant embryos(Fig. 4L). In a control experiment, shot³/+; kra²/+ embryos showed no significant midline defects (Fig. 4L). These dosage-sensitive genetic interactions, coupled with their biochemical interactions, suggest that Kra and Shot function together to control midline axon repulsion.

The phenotype observed in kra and shot mutants is qualitatively identical to that seen in robo mutants(Kidd et al., 1998),suggesting a link between Kra/Shot and the Slit/Robo repellent pathway. To test this hypothesis, we examined transheterozygous interactions of kra and shot with robo. In a control experiment,neither kra²/+ nor shot³/+ embryos showed significant axonal defects at the midline(Fig. 5A,B). In embryos heterozygous for robo², we rarely observed ectopic midline crossing of Fas II-positive axons (Fig. 5C). However, the phenotype was dramatically enhanced in transheterozygous robo²/+; kra²/+ or shot³,robo²/+ embryos (>17-fold; Fig. 5D,E,G). The phenotype was further increased in embryos transheterozygous for shot, kra and robo (Fig. 5F,G). Similar genetic interactions were also observed between kra, shot and slit (see Fig. S2 in the supplementary material). Thus, kraand shot genetically interact with robo and slit,suggesting that all these genes may function in the same guidance process.

To further define the role of Kra in midline axon repulsion, we performed detailed phenotypic analyses of kra mutant embryos. In the embryonic CNS, glial cells provide neurons with cues and substrata for growth cone migrations (Chotard and Salecker,2004). Therefore, the axon phenotype observed in kracould, in principle, be because of defects in CNS glia. This possibility led us to examine kra mutant embryos for glial defects. For this purpose,we also examined the axonal trajectories of the Ap-expressing neurons by introducing an apC-tau-lacZ transgene(Lundgren et al., 1995) into the wild-type and kra mutant backgrounds. Because Ap axons express Robo (Rajagopalan et al.,2000b), they provide an excellent opportunity to investigate the midline behaviors of Robo-positive axons at single axon resolution. In stage 16 wild-type embryos, the Ap axons remained ipsilateral without crossing the midline (see Fig. S3A,B in the supplementary material)(Lundgren et al., 1995). However, in stage 16 kra¹/kra² embryos, the Ap axons ectopically crossed the midline in 13% of segments despite the apparently normal arrangement of glial cells as defined by anti-Repo and anti-Wrapper (see Fig. S3C,D in the supplementary material). Thus, the ectopic midline crossing of CNS axons in kra mutant embryos cannot be ascribed to gross defects in the CNS glia.

We then examined the expression of the midline repellent Slit and its Robo receptor. In stage 16 wild-type embryos, Slit protein was detected at high levels around the midline glia (Fig. 6A,C), whereas Robo was highly expressed on the longitudinal axon tracts and largely absent from the commissural tracts(Fig. 6G,I)(Kidd et al., 1999). In the same embryos, the Ap axons remained ipsilateral within the Robo-positive longitudinal tracts (Fig. 6B,C,H,I). However, in stage 16 kra¹/kra² embryos, the Ap axons ectopically crossed the midline despite normal Slit expression along the midline(Fig. 6D-F). Surprisingly, Robo protein was aberrantly detected in some commissural axon tracts(Fig. 6J). The commissural tracts expressing Robo frequently contained Ap axons that were ectopically crossing the midline (Fig. 6K,L), suggesting that these axons contribute to aberrant Robo expression. By contrast, Robo expression in the longitudinal axon tracts of kra mutant embryos was approximately the same as that of wild-type embryos (Fig. 6G,J). Taken together, these data support an essential role of Kra in maintaining the sensitivity of Robo-expressing growth cones to the midline repellent Slit. However, we cannot exclude the possibility that Robo misexpression by axons normally crossing the midline also contributes to aberrant Robo expression in kra mutant embryos.

All other proteins containing the W2 domain are translation initiation factors and associate with 40S ribosomal subunits(Preiss and Hentze, 2003). To determine whether Kra also associates with 40S ribosomal subunits, Drosophila S2 cells were treated with cycloheximide to arrest translation, and their extracts were fractionated on sucrose gradients. Kra cosedimented with the mRNA cap-binding protein eIF4E, but not with the ribosomal protein L28, suggesting that it is associated with 40S, but not 60S,subunits (Fig. 7A).

The W2 domains of eIF5 and eIF2Bϵ bind directly to the initiation factor eIF2β (Asano et al.,1999), the β-subunit of eIF2 whose GTPase activity is essential for 40S and 60S subunits joining into 80S complexes. eIF5 is a GTPase activating protein (GAP), and eIF2Bϵ is a GDP exchange factor(GEF) for eIF2. We therefore asked whether Kra also binds to eIF2βthrough its W2 domain. We found that GST-Kra binds to eIF2β in vitro(Fig. 7B). We observed this interaction in the fly embryo as well (Fig. 7C). Both 12A and 7A mutations significantly impaired Kra binding to eIF2β (Fig. 7C),suggesting that the binding of Kra to eIF2β requires the residues in its W2 domain previously identified as necessary for eIF5 binding to eIF2β(Asano et al., 1999). Because the same W2 domain also binds to Shot, we tested whether the formation of a multiprotein complex containing Shot, Kra and eIF2β is possible. In S2 cells, C-Shot L-GFP coimmunoprecipitated with endogenous Kra and eIF2β(Fig. 7D). As C-Shot L does not directly bind to eIF2β (data not shown), this result suggests that Kra can simultaneously bind to Shot and eIF2β in vivo.

Our results described above suggested that Kra may compete with eIF5 and eIF2Bϵ for eIF2β. We therefore tested whether Kra inhibits translation in vitro. Translation of luciferase mRNA in reticulocyte lysate was specifically inhibited by Kra, but not by BSA and GST control proteins (Fig. 7E). Importantly, we found that these proteins do not affect the stability of luciferase mRNA (Fig. 7E). Translation inhibition by Kra was dose-dependent at concentrations ranging from 240 nM to 2.4 μM(Fig. 7E). Interestingly, the 7A mutation significantly impairs Kra activity to inhibit translation(Fig. 7E), suggesting that protein-protein interactions through the AA-box 2 of Kra may be essential for its activity to inhibit translation. The extreme insolubility of Kra-12A in E. coli or insect cells prevented us from including it in this assay.

Our biochemical data suggested that the AA-boxes in the W2 domain of Kra are crucial for the formation of complexes containing Shot and eIF2β. To directly test the importance of these protein-protein interaction domains in axon guidance, we tested whether mutations in the AA-boxes would impair the ability of Kra to rescue kra mutant phenotypes in the embryonic CNS. Neuronal expression of Kra12A or Kra-7A failed to rescue the midline guidance defect in kra¹/kra² embryos(Fig. 8). The expression levels and solubility of Kra-12A and Kra-7A were comparable to those observed for the wild-type Kra (data not shown). Thus, the ability of Kra to form protein complexes with Shot and eIF2β is essential for its function in midline axon repulsion.

To confirm the implication of the Kra binding partner eIF2β in midline axon repulsion, we looked at CNS axons in eIF2β mutant embryos. We observed ectopic midline crossing by the Ap axons and axons from the pCC ipsilateral pathway (see Fig. S4 in the supplementary material), which is reminiscent of the kra loss-of-function phenotypes. This observation further supports a role for eIF2β in Kra-mediated midline axon repulsion and provides in vivo evidence for the requirement of protein translation in axon guidance.

Kra and its human homolog BZAP45 contain an N-terminal leucine-zipper domain of unknown function and a C-terminal W2 domain. We show here that Kra can bind to eIF2β through its W2 domain and inhibit translation in vitro. It is very likely that Kra competes with eIF5 and eIF2Bϵ for the common binding partner eIF2β, thus inhibiting the assembly of functional preinitiation complexes. A similar mode of translation inhibition has been proposed for DAP-5/p97, which may compete with its homolog eIF4G for eIF3 and eIF4A, thus reducing both cap-dependent and -independent translation(Imataka et al., 1997; Yamanaka et al., 1997). However, the step in translation initiation that is regulated by Kra remains to be addressed experimentally.

Kra-mediated translational repression appears to be an important mechanism underlying midline axon guidance. In the kra mutant embryos, Fas II-positive CNS axons that normally remain ipsilateral cross the midline ectopically. This phenotype is observed with the pCC axons from early stages(stages 12 and 13) of axogenesis when they pioneer one of the Fas II pathways. The introduction of multiple alanine substitutions (12A and 7A) into Kra significantly reduces its ability to bind eIF2β and abolishes its activity to rescue the kra mutant phenotype, suggesting that the function of Kra in axon guidance depends on its interaction with eIF2β. Consistent with this conclusion, mutations in the eIF2β gene also lead to the ectopic midline crossing of Fas II-positive axons.

There is a growing body of evidence that F-actin and microtubules are coordinately assembled to each other during axon extension and guidance(Kalil and Dent, 2005; Schaefer et al., 2002; Zhou et al., 2002). Interactions of filopodial actin bundles and microtubules are key features of filopodial maturation into an axon (Sabry et al., 1991) and of growth cone turning(Zhou et al., 2002). Shot, a conserved molecule that scaffolds F-actin, microtubules and the microtubule plus end-binding protein EB1 (Lee and Kolodziej, 2002b; Subramanian et al., 2003), is a strong candidate to bring microtubule plus ends into contact with F-actin bundles. Indeed, Shot is required for the extension of sensory and motor axons (Lee and Kolodziej, 2002b), and a mammalian homolog of Shot, ACF7, is required for microtubules to track along F-actin cables towards the leading edge of spreading endodermal cells (Kodama et al., 2003). Thus, previous studies have suggested that Shot/ACF7 coordinately organizes F-actin and microtubules to support the motility of neuronal growth cones and nonneuronal cells.

Our findings suggest that Shot also functions together with the translation inhibitor Kra to control midline axon repulsion. Shot physically associates with Kra in vivo. The shot loss-of-function phenotype at the CNS midline is reminiscent of the kra loss-of-function phenotype. The major Kra-binding domain in Shot is required for its role in midline axon repulsion. Moreover, shot and kra genetically interact in a dosage-sensitive manner for the midline phenotype. Our data also support the idea that cytoskeletal assembly and translational regulation can occur in a coordinated way. We found that midline axon repulsion requires both the activity of Kra to recruit eIF2β and the activity of Shot to bind to F-actin. Thus, it is likely that local levels of eIF2β available for protein synthesis can be spatially regulated with regard to actin cytoskeleton remodeling during axon guidance.

In Drosophila, Slit is the key ligand driving midline axon repulsion (Kidd et al., 1999),and therefore midline crossing of CNS growth cones is primarily controlled by the Robo receptor of Slit (Rajagopalan et al., 2000a). How then do neurons regulate levels of Robo on the surface of their axons and growth cones? The transmembrane protein Commissureless (Comm) has been shown to dynamically regulate Robo expression(Keleman et al., 2002; Myat et al., 2002). Comm functions as an intracellular sorting receptor to target newly made Robo for lysosomal degradation, thereby blocking its transport to the growth cone that is crossing the midline (Keleman et al.,2002; Keleman et al.,2005).

Translational regulation has also been shown to alter the responsiveness of growth cones to the midline repellents(Dickson, 2002). In vitro studies of cultured embryonic retinal ganglion cells (RGCs) provided an insight into how translation is regulated in axons and growth cones in response to midline guidance cues. Treatment of these neurons with netrin-1 leads to the rapid activation of signaling pathways that phosphorylate the translation initiation factor eIF4E and its binding protein eIF4E-BP1 and thus induces axonal protein synthesis (Campbell and Holt, 2001; Campbell and Holt, 2003). Our data presented here indicate that the role of Kra in midline axon repulsion depends on its ability to recruit the translation initiation factor eIF2β to Shot. Thus, protein complexes containing Shot,Kra and eIF2β may function as additional targets for signaling systems that critically control axon guidance at the CNS midline. Regulation of Shot-Kra-eIF2β complexes may occur in neuronal cell bodies, where Kra is concentrated, or in axons and growth cones, which may require local protein synthesis to meet developmental requirements. In the latter case, as Kra is not detectable in the CNS axons of the Drosophila embryo, even a low amount of Kra may be sufficient for guiding axons.

Intriguingly, Robo was aberrantly detected on commissural axons in kra¹/kra² mutant embryos. Given the increased frequency of ectopic crossovers in these embryos, as well as the documented role of Robo in preventing axons from crossing the midline(Kidd et al., 1998), this finding is somewhat paradoxical. One possibility is that Kra, induced by interaction with Shot and eIF2β, could repress the synthesis of as yet unidentified proteins that transduce or modulate Robo signaling. In shot or kra mutant embryos, perhaps this translational regulatory circuit is not activated, and thus Robo-expressing growth cones abnormally cross the midline because of a decrease in the strength of Robo signaling output. Alternatively, Kra may function to finely tune the expression levels of their multiple targets that mediate attractive or repulsive responses. In this scenario, impairment of the Shot-Kra-eIF2βcircuit could disturb the precise balance between repulsion and attraction signaling at the midline, thereby decreasing the overall sensitivity of Robo-expressing growth cones to Slit. Therefore, efforts to reveal the direct targets of Kra-mediated repression in the future may provide better insights into the immediate mechanisms by which translational regulation plays an essential role for midline axon repulsion.

We thank Haejung Won for her able technical assistance, and Changjoon Justin Lee, Kei Cho and Andy Furley for their valuable comments. We thank Akira Nakamura for the anti-eIF4E antibody, Edward Giniger for the apC-tau-lacZ transgenic line, Sang-Hak Jeon for the repo-GAL4 line and Christian Klämbt for the slit-GAL4line. This work was supported by funds from the Brain Research Center of the 21st Century Frontier, the Basic Research Program of the Korea Science and Engineering Foundation (R01-2006-000-10487-0) and the Korea Research Foundation (KRF-2003-015-C00472 and KRF-2006-0409-20060081).