It has been previously shown that Slit and the three Drosophila Robo receptors are expressed in the medulla neuropil in addition to other regions of the visual system (Tayler et al., 2004; Suzuki et al., 2016; Caipo et al., 2020; Guzman-Palma et al., 2021). Furthermore, in our previous work, we demonstrated that Ey + medulla neurons in the optic lobe are an important source of Slit (Caipo et al., 2020). To further characterize the expression of Slit and Robo receptors in the optic lobe, we performed immunostaining of the four proteins in the L3 stage and examined horizontal sections of the optic lobe, which allowed us to observe all neuropils in the same plane (Figures 1A-M). Interestingly, we noted that the Slit signal is detected in the same regions in which Robo receptors are expressed, suggesting that Slit and Robo receptors are co-expressed in the optic lobe (Figures 1D,G,J,M).

The null mutant allele of the robo3 gene, robo3³ (Supplementary Figures S1A–D), has been previously characterized; it showed similar phenotypes to slit mutants in photoreceptor axons (Pappu et al., 2011). Due to the resemblance in the mutant phenotypes, we decided to study the relationship between Slit and Robo3 in medulla neurons. Examination of Slit and Robo3 expression in Ey + medulla neurons indicate that these two proteins co-localize in Ey + medulla neurons in a punctate pattern, especially in projections and the plexus region where growth cones are located (Figure 2 A–D, a`–a` ` `). In addition, using a Slit-GFP reporter (Figures 2E–e` `) (Plazaola-Sasieta et al., 2019; Caipo et al., 2020) to identify Slit-expressing cells, we confirmed that cells expressing Slit in the medulla neurons are also positive for Robo3 expression. These results support that Slit and Robo3 are co-expressed in medulla neurons, suggesting the possibility of an autocrine/paracrine pathway.

One of the most prominent phenotypes of slit mutants in the optic lobe is the intermingling of lobula complex cells with lamina and medulla neuropils during development (Tayler et al., 2004; Suzuki et al., 2016; Plazaola-Sasieta et al., 2019; Caipo et al., 2020), which leads to strong morphological defects in the adult optic lobe. A similar phenotype is observed when Slit is knocked down, specifically in Ey + medulla neurons (Caipo et al., 2020). In contrast to Slit, the cell-specific requirements of Robo receptors in the optic lobe are less characterized, although it was previously shown that knocking down the three Robo receptors simultaneously in all neurons produced a phenocopy of the slit mutant (Tayler et al., 2004). Studies in whole mutant animals showed that robo2 and robo3 mutants at the larval stage displayed similar boundary defects to those observed in slit mutants (Suzuki et al., 2016). In addition, we previously assessed the phenotypes of robo2 mutants in the adult optic lobe, finding that they are subtler than the phenotypes of slit mutants in adult animals, and connected to their function in the lobula plate (Guzman-Palma et al., 2021). In the case of Robo3, we examined the adult optic lobe in robo3³ mutants (Figures 3A–D) and noticed defects in neuropil organization that are similar to those observed in the sli ^dui mutant, which is a hypomorphic allele with decreased Slit expression, especially in the optic lobe (Figure 3E and Supplementary Figures S1E–H). In robo3³ mutants, we observed strong medulla defects, in addition to the R-cell defects already reported. These results suggest that Robo3 plays a critical role in the development of the optic lobe particularly in the medulla. Next, we decided to test whether Robo3 or the other Robo receptors have autonomous roles in medulla neurons. We performed knockdowns using shRNAi with the GAL4/UAS system (Brand and Perrimon, 1993) in Ey + medulla neurons using the ey ^OK107 -GAL4 driver (Figures 3F–J) expressing previously tested RNAi lines in the optic lobe ((Guzman-Palma et al., 2021) Supplementary Figures S1I–Y) and analyzed the morphology of the adult optic lobe. We found that only the Robo3-RNAi knockdown (KD) produced strong alterations in optic lobe morphology, using two independent lines (Figures 3I,J), which is a consequence of the compartmentalization defect produced during development. These alterations are similar to those of slit and robo3 mutants and to the Slit-KD in medulla neurons reported previously (Figures 3C-E) (Caipo et al., 2020). To better characterize the role of Robo3 in optic lobe development, we performed Robo3 KD experiments in photoreceptor and glial cells. We found that Robo3 is also important in photoreceptor cells since Robo3 KD led to strong medulla defects (Supplementary Figure S2A–D) and in glial cells where Robo3 KD led to subtler defects in photoreceptor axons (Supplementary Figure S2E–H). Notably, we also detected Robo3 expression in glial cells sitting on the lamina plexus (Supplementary Figure S3A–F). These results suggest that Robo3 is required in several cell populations for correct optic lobe compartmentalization.

In summary, our data show that Slit and Robo3 are co-expressed and required for the development of medulla neuron, and suggest that the mechanism involves autocrine/paracrine signaling at least in Ey + medulla neurons.

In recent years, several cellular mechanisms have been shown to collaborate with Slit–Robo for signal transduction. It has been reported that Slit-Robo1 signaling in the VNC of the Drosophila embryo is regulated by endocytosis (Chance and Bashaw, 2015). Interestingly, Robo3 also bears a predicted putative motif for clathrin-dependent endocytosis (not shown). Therefore, we assessed whether endocytosis and the post-endocytic trafficking play a role in the Slit/Robo3 autocrine/paracrine pathway using Airyscan confocal microscopy. First, we expressed GFP reporters for main Rab GTPases located in different types of endosomes (Zerial and McBride, 2001) using the ey ^OK107 -GAL4 driver (Figures 4A–C). We used Rab5 (early endosomes), Rab7 (late endosomes), and Rab11 (recycling endosomes). Importantly, the expression of these fusion proteins did not alter the morphology of the optic lobe (Supplementary Figure S4). We found that Slit and Robo3 are localized in all types of endosomes, especially in Rab5 and Rab11 positive endosomes and to a lesser degree in Rab7 positive endosomes (Figures 4D-O). However, the presence of Slit and Robo3 in endosomes could originate from anterograde trafficking after the process of protein synthesis. To confirm that sorting to endosomes can occur from the cell membrane upon endocytosis, we performed internalization assays using primary cell cultures of larval brains (Figures 5A–J) expressing the different Rab reporters under the control of ey ^OK107 -GAL4. Cells were incubated with Slit-myc-Cherry obtained from a stably transfected Drosophila S2 cell line (Supplementary Figure S5). The Slit-myc-Cherry protein can be detected in S2 cell media (Supplementary Figure S5A–C), and its overexpression can disturb the development of photoreceptor axons (Supplementary Figure S6A). However, Slit-myc-Cherry expression in Ey + neurons could not rescue the slit mutant phenotype in the optic lobe (Supplementary Figure S6B), as previously observed for an untagged Slit protein (Caipo et al., 2020). These results indicate that this tagged version is less active or may have a lower rate of synthesis/secretion. We observed Robo3 in early (Rab5+), late (Rab7+), and recycling (Rab11+) endosomes 15 min after treatment with Slit (Figures 5G–I). Furthermore, incubation with Slit increased the co-localization of Robo3 receptors to late endosomes compared to the mock treatment (Figures 5A–D,G–H). Finally, we tested whether N-Slit (the Slit fragment that contains the Robo binding site) moves to endosomes from the cell membrane upon incubation with neurons. We observed that Slit is present in the three types of endosomes 15 min after incubation (Supplementary Figure S7 and Supplementary Figure S8). These results indicate that Slit and Robo3 can be endocytosed to enter the recycling route (Figure 5J).

We performed genetic interaction experiments, in which RNAi or dominant-negative proteins for distinct components of the endocytic machinery were expressed in Ey + medulla neurons combined with heterozygotes for slit or robo3 mutants (Figures 6A-R, Supplementary Table S1). We included an ey ^3.5 -GAL80 transgene to avoid the expression of the driver in the eye-antennal imaginal disc (Supplementary Figure S9), which could affect optic lobe development (Huang and Kunes, 1996). For these experiments, we focused on two qualitatively distinct phenotypes: the appearance of strong defects in the medulla neuropil, in which the medulla is disrupted or disorganized, and the alterations in photoreceptor axons, such as ectopic photoreceptors (see Supplementary Figure S10 for a detailed description), which are likely to be indicators of more subtle defects in medulla organization and/or defects in Slit availability in the extracellular milieu. In addition, we observed axonal swelling defects in photoreceptor axons when Clathrin was disturbed, which may reflect its involvement in additional signaling pathways. Therefore, we did not consider this phenotype in our quantification analysis (Supplementary Figure S10 N–O ` ` `)). We found strong genetic interactions of slit and robo3 mutants with clathrin and rab11 and milder defects with rab5 and rab7 (Figures 6Q–R, Supplementary Table S1). Thus, our results suggest that the Slit-Robo3 signaling pathway involves the participation of the endocytic machinery, including Rab11 and recycling endosomes. These data contrast with a previous study reporting strong interactions of Slit-Robo1 with Rab7 suggesting the participation of late endosomes (Chance and Bashaw, 2015).

After establishing the participation of the endocytic pathway in Slit-Robo3 signaling, we assessed whether classical downstream targets of Robo1 in other contexts also take part in the process of medulla development (O'Donnell et al., 2009). We performed genetic interaction experiments with dominant-negative forms of the small GTPases Rac, RhoA, and Cdc42, and we scored the same phenotypes described before (see Supplementary Figure S10 for a detailed description). We found a strong genetic interaction of Slit-Robo3 with Rac1 and Cdc42 (Figures 7A-P, Supplementary Table S1), suggesting that these proteins are the main downstream effectors in this context.

The general concept of axon guidance cues in the nervous system is that a high concentration of the cue in one region will instruct the navigation of axons according to the repertoire of receptors in their membranes. However, we propose that the Slit gradient is not required for optic lobe morphogenesis. This work, as well as our previous work and reports from other labs, present several arguments in favor of this idea: 1) there is no gradient observed in the larval optic lobe; 2) the slit mutant phenotypes can be rescued by expressing Slit in various cell populations, such as medulla neurons and, partially, in glia and photoreceptor cells (Tayler et al., 2004; Caipo et al., 2020); and 3) large loss of function clones carrying the slit ² allele in the visual system do not produce defects in optic lobe organization (Tayler et al., 2004). Indeed, Slit expression does not show a restricted source in the larval optic lobe, and previous data show that it is expressed in multiple cell types (Tayler et al., 2004; Caipo et al., 2020; Guzman-Palma et al., 2021), unlike in the central brain where it is enriched in the mushroom body (Oliva et al., 2016) and the VNC where it is expressed in the midline (Kidd et al., 1999; Dickson and Gilestro, 2006). Regarding the function of Robo receptors, the knockdown of all three Robo paralogues using a general driver resulted in defects in optic lobe development (Tayler et al., 2004). Here, we showed that the Robo3 function is required in medulla neurons since RNAi knockdown produces similar phenotypes as those observed in the robo3 mutants. Robo3 is expressed in R8 photoreceptors where it regulates axon guidance, in response to Slit expressed in the optic lobe (Pappu et al., 2011). Thus, Slit performs two tasks during the development of the visual system. It serves as an axon guidance cue for photoreceptor axons (and perhaps other neurons) and it also prevents the intermingling of neighboring cell populations in the optic lobe.

In the nervous system, Slit and Robo receptors are largely expressed in complementary patterns (Kidd et al., 1998; Brose et al., 1999; Kidd et al., 1999; Dickson and Gilestro, 2006). Thus, in most contexts, Slit is secreted by a discrete group of cells, playing instructive roles for surrounding axons. Classic examples are the midline glia in insects and the floor plate in vertebrates, which play homologous functions in guiding commissural axons (Kidd et al., 1998; Brose et al., 1999; Kidd et al., 1999; Dickson and Gilestro, 2006). Few examples of Slit–Robo signaling playing an autocrine/juxtracrine role in the nervous system have been reported. In mouse, it can promote fasciculation of motor neurons (Jaworski and Tessier-Lavigne, 2012). In this case, Slit secreted by motor neurons binds to Robo1 and Robo2 in axons; it is necessary to avoid premature defasciculation at muscle targets. Another example is the co-expression of Slit2 and Robo2 during the development of Purkinje cells (Gibson et al., 2014). The deletion of either of these proteins leads to excessive dendrite self-crossing, demonstrating the role of this pathway in self-avoidance. One question is whether the downstream effectors are the same in these contexts. Regarding fasciculation, it is likely that the downstream effectors are adhesion molecules, such as cadherins, which are regulated by Robo in several different contexts (Rhee et al., 2002; Rhee et al., 2007; Guzman-Palma et al., 2021). Here, we find that in this context, regulators of the actin cytoskeleton are likely to be the downstream effectors of Robo3 in the optic lobe, although more work is required to unravel the complete mechanism.

The current model of neuropil compartmentalization in the optic lobe largely depends on the interplay between cell repulsion and attraction (Suzuki et al., 2018), in which complementary populations express ligands and receptors. Our data shows that this view has to be complemented based on the co-expression of Slit and Robo in at least some cellular populations. One interesting possibility is that autocrine/paracrine Slit-Robo3 signaling regulates the formation of actomyosin fibers that restrict cell movement from neighboring neuropils, leading to tissue separation. This mechanism has been extensively documented in vertebrates (Kiecker and Lumsden, 2005; Calzolari et al., 2014; Addison and Wilkinson, 2016) and also observed in other Drosophila tissues during development (Monier et al., 2010). RhoA is generally recognized as the GTPase that promotes the formation of actomyosin fibers. As shown here, recent literature also supports the participation of Rac and Cdc42 (which interact genetically with Slit in optic lobe development) in the initial steps of actomyosin fiber formation (Arnold et al., 2017). Another possibility is the regulation of cell repulsion, in which Slit binds Robo3 in the medulla and other Robo receptors (or a combination of them) in the lobula complex and lamina, which can promote disruption of cell–cell contacts at the interface between neuropils. This mechanism could be similar to the one involved in the separation of ectoderm and mesoderm in the early frog embryo, which depends on two antiparallel Eph-ephrin signaling processes (Wilkinson, 2021) triggered by both tissues (Rohani et al., 2011). Interestingly, one of the downstream effectors found in this study is Rac1, which can rescue the absence of Eph-Ephrin signaling in these tissues. One point in favor of this alternative is its compatibility with developing axons and dendrites undergoing migration in both neuropils using one another as substrate, which also happens in the interface between ectoderm and the migrating cells from the mesoderm.

The role of endocytosis is currently recognized as an important factor in regulating several signaling pathways (Bokel and Brand, 2014; Cosker and Segal, 2014). In the case of the Slit–Robo pathway, only a few publications have explored this aspect during the development of the fly nervous system. Chance and Bashaw (2015) found that endocytosis was important for the function of Slit-Robo1 in VNC axon guidance in the fly embryo. In contrast to our data on Robo3, the authors found strong genetic interactions with mutants of Rab7 indicating a major role of the late endosomes, while we found that Rab11 and presumably recycling endosomes, may have a main role in optic lobe development. Resensitization could explain the importance of recycling, in which receptors are trafficked back to the cell membrane (Roosterman et al., 2004; Hinkle et al., 2012; Kharitidi et al., 2015). This explanation is in line with results reported in mice in which the GTPase Arf6 regulates Robo1 membrane availability to increase the repulsion of post-crossing axons during spinal cord development (Kinoshita-Kawada et al., 2019). Arf6 promotes the sorting of Robo1 to recycling endosomes, and the authors also observed functional interactions with Rab11 proteins in their primary culture experiments. Why is recycling important in the case of the optic lobe but not as crucial in the midline as shown by Chance and Bashaw? A possible explanation is that the levels of Slit in the optic lobe are lower than those in the embryo VNC (see Tayler et al., 2004, Figure 4A and Oliva et al., 2016; Figures 1D,E). Therefore, a continuous supply of receptors to the membrane may be required for adequate levels of signaling. Interestingly, the transfer of receptors to recycling endosomes can be favored by a low concentration of the ligand in some systems (Roosterman et al., 2004).

Since Robo receptors are interchangeable in the optic lobe (Pappu et al., 2011), the downstream target activated in this context must share signaling molecules. The actin cytoskeleton is a common target for Slit–Robo signaling, which is conserved across evolution. Regulators of actin polymerization such as small Rho GTPases are downstream of Robo receptors in C. elegans, flies, and vertebrates (O'Donnell et al., 2009).

The cytoplasmic tails of all Robos have some conserved regions known as the CC domains. Drosophila Robo1 has four of these domains (CC0-3) but Robo2 and Robo3 have only CC0-CC1(Dickson and Gilestro, 2006). Most of the identified downstream effectors of Robo1 have been described. However, little is known about Robo2 and Robo3 effectors. For instance, several Robo1 effectors are recruited using CC2 and CC3 domains, such as SOS and Pak, which can activate Rac1 and Cdc42 (Fan et al., 2003; Yang and Bashaw, 2006). Since Robo3 is lacking CC2-3, it is still unclear how it may activate Rac1 and Cdc42, and therefore more work is required in this direction.

Flies were raised on standard fly food at 25°C for an expression pattern analysis, genetic interaction, Rabs-GFP phenotype, and co-localization experiment. Slit-myc-Cherry overexpression and knockdown experiments using RNAi were performed at 29°C. UAS-Slit-myc-Cherry expressing flies were generated by BestGene Inc., United States. All other lines were obtained or generated using fly strains from the Bloomington Drosophila Stock Center (Bloomington, Indiana). The details of genotypes used throughout this work are presented in Supplementary Table S2.

The Slit isoform C was used (FlyBase) for the design of the Slit-myc-Cherry construct. A sequence coding for a myc epitope (EQKLISEEDL), flanked by two Ig2 linker sequences (IASKPKGASVRA), was inserted at the end of the fifth EGF-like repeat, before the cleavage site (PDDYTGKYCEGHNMISMMYPQTSP). The stop codon was removed and the sequence coding for mCherry was inserted after an Ig2 linker sequence. The complete sequence was codon-optimized and KpnI restriction sites were added flanking the sequence. The construct was synthesized and cloned into a pUAST-attb vector in the KpnI site by GENEWIZ Inc., United States.

Larval L3 and adult stage brains were dissected using standard procedures (Wu and Luo, 2006). In brief, brains were dissected and fixed in 4% formaldehyde in 1X PBS for 20 min at room temperature. Then, samples were washed six times in PBT (0.3% Triton X-100 in 1X PBS) and blocked in 1% NGS/PBT for 30 min at room temperature. Primary antibodies were diluted in PBT and samples were incubated overnight at 4C. The next day, samples were washed six times with PBT, incubated with fluorescent-dye conjugated secondary antibodies for 2 hrs, and washed six times in PBT. Then, samples were incubated with Hoechst (Thermo Fisher Scientific) in PBT for 10 min and washed three times with PBT. Finally, samples were incubated for 1 h in 50% glycerol/PBS at 4°C. Samples were mounted using Vectashield mounting medium (Vector). Primary monoclonal antibodies obtained from Developmental Studies Hybridoma Bank are mouse anti-Slit (C55.6D; 1:10), mouse anti-Robo1 (13c9; 1:50), mouse anti-Robo3 (14c9, 1:50), mouse anti-Chaoptin (24B10; 1:10), and rat anti N-Cadherin (DN-Ex #8; 1:10). Other antibodies used were mouse anti-myc (9e10; Santa Cruz 1:100), rabbit anti-HA (c29f4; Cell Signaling 1:100)), rabbit anti-GFP (A6455 Invitrogen, 1:100), and anti-Cherry (632,543, ClonTech Laboratories; 1:1,000). Fluorescent-dye conjugated secondary antibodies were obtained from Jackson Immunoresearch (Pennsylvania, United States) and Invitrogen (Massachusetts, United States) and used 1:200.

For expression pattern analyses, images of the larval brain were acquired using a Z-step size of 1.5 μm. Adult optic lobe stacks were acquired using a Z-step size of 0.8 μm. For adult optic lobes, the images shown are Z projections or a single slice (indicated in figure legends). Images of larval primary cell cultures were acquired using a Z-step size of 0.5 μm. All images were acquired using an Airyscan confocal microscope (Zeiss) at the UMA PUC facility.

Co-localization analyses were performed using the Jacob plugin of Fiji in which Manders M1 represents channel 1 (Rabs) co-localized with channel 2 (Slit or Robo3). For in vivo co-localization of Rab proteins with Robo3 and Slit (n = 5 optic lobes), stacks with a Z-step size of 0.5 μm were acquired. For in vitro experiments, co-localization of Rab5 with Robo3 (n = 5) and Rab7 and Rab11 (both n = 4), 12–21 cells were used per sample. For co-localization of Rab proteins with N-Slit (anti-myc) 7–10 cells were used per sample (n = 2 for Rab5; n = 3 for Rab7 and Rab11). For in vitro experiments using cell culture, stacks were acquired with a Z-step size of 0.5 μm, and the representative images were generated using a Z projection of two sections in the middle of the stack.

For co-localization analyses of Slit and Robo3-HA (endogenously tagged) in the larval stage n = 5, Slit was acquired in channel 1 and HA in channel 2, and for graphs Manders M2 value was used. For co-localization experiments, a statistical analysis was performed using the Mann–Whitney non-parametric test, using Prism 8 software.

For a phenotypic analysis of mutant and RNAi-expressing animals, the sample sizes were robo3³ : n = 10 and Robo KD: n = 10. In the case of genetic interaction experiments, n = 15 in all experimental conditions (an independent set of control flies was used in each experiment). The evaluated phenotypes of genetic interaction experiments are schematized in Supplementary Figure S10. For Rab proteins, overexpression experiments, n = 7; overexpression of Slit-myc-Cherry in adult stage using GMR-GAL4, n = 10.

Stable S2 cells transfection was performed following the instruction of Thermo Fisher Scientific Manual 0000656 rec B0 catR69007. Four days after transfection, culture medium was replaced by Schneider insect medium supplemented with 7% fetal bovine serum (7% FBS, Biological Industries) and 300 μg/ml hygromycin (1,068,701, Invitrogen) and plated in a 96 well plate. After 3 days, the best clone was selected using an epifluorescence microscope to start its expansion. S2-transfected cells with mock (including pUAST empty vector + actin-GAL4 +pCoHygro (resistance vector)) or Slit-myc-Cherry (including pUAST-Slit-myc-Cherry + actin-GAL4 +pCoHygro) were plated using the Schneider medium. The Slit-myc-Cherry or mock conditioned mediums was collected 48 h after plating the medium, and this step was repeated three times every 24 h. The medium was centrifugated to remove cells and concentrated using an Amicon ultra-15 of 100KDA filter (Millipore). Western blot assays were performed to confirm the presence of Slit-myc-Cherry in the S2 cells and the conditioned medium. S2 cells were lysed using 100 μl of lysis buffer (20 nM HEPES, pH7.5, 100mM KCl, 5% glycerol, 10 mM EDTA, 0.1% Triton X-100, 1 mM DTT, and protease inhibitors). Samples of S2 mock and Slit-myc-Cherry (50 µg of total protein) and 10 µl of the concentrate conditioned medium were heated at 95°C for 5–10°min and loaded in 7.5% SDS-PAGE gels. The membrane was incubated overnight at 4°C with mouse anti-myc antibody 1:500 (Santa Cruz) diluted in blocking solution (5% milk in 0.1% Tween 20 PBS 1X). The secondary antibody was incubated for 2 h at room temperature in the blocking solution. A chemiluminescence reaction was performed using a WESTAR Supernova reagent (XLS3,0100 Cyanagen) and acquired using the UVITEC imaging system.

A primary cell culture was performed according to (Sicaeros et al., 2007) but using the larval L3 developmental stage. Larvae were rinsed with 70% ethanol followed by two more rinses in distilled water. Larval brains were dissected in dissecting solution (DS: 6.85 mM NaCl Na₂HPO₄, 0.001 Mm KH₂PO₄, 0.2772 mM HEPES, pH 7.4), and briefly spun. Then, brains were treated with papain (LS 03126, Worthington) for 30 min at room temperature. Samples were washed three times with a DMEM/F12 culture medium (12,400–016, Gibco) supplemented with 100 μg/ml Apo-transferrin, 30 nM selenium, 50 μg/ml insulin, 2.1 mM 20-hydroecdysone, 20 ng/ml progesterone, 100 μM putrescine (all from Sigma Aldrich), and 1% antibiotic/antimycotic (15,240,062, Gibco). The tissues were then mechanically disaggregated and mounted on Laminin/Concavalin-coated coverslips in the presence of DMEM/F12-supplemented medium. Cells obtained from two larval brains were used for each coverslip. The following day, conditioned media obtained from astrocytes cultured in the neurobasal medium supplemented with B27 (CNBM/27) was added to the cells. On the fifth day in vitro of the experiment, samples were processed for internalization assays.

Primary cells culture of eyeless expressing neurons, bearing Rabs-GFP transgenes were incubated in 50% of mock CM or Slit-myc-Cherry CM for 15 min at 25°C. Cells were washed with DMEM/F12-supplemented medium at 4°C and then with acidic pH 3.6. Then, cells were fixed with 2% formaldehyde in 4% sucrose–PBS 1X for 20°min, followed by incubation with 0.15 M glycine for 15 min. For immunofluorescence, the primary antibodies used were mouse anti-Robo3 or rabbit anti-myc. The secondary antibody was obtained from Invitrogen (used 1:200) and the mounting medium used was Fluoromount-G™ (17,984–25, electron microscopy sciences).