|Home | About | Journals | Submit | Contact Us | Français|
How neurons form synapses within specific layers remains poorly understood. In the Drosophila medulla, neurons target to discrete layers in a precise fashion. Here we demonstrate that the targeting of L3 neurons to a specific layer occurs in two steps. Initially, L3 growth cones project to a common domain in the outer medulla, overlapping with the growth cones of other neurons destined for a different layer, through the redundant functions of N-Cadherin (CadN) and Semaphorin-1a (Sema-1a). CadN mediates adhesion within the domain and Sema-1a mediates repulsion through Plexin A (PlexA) expressed in an adjacent region. Subsequently, L3 growth cones segregate from the domain into their target layer in part through Sema-1a/PlexA-dependent remodeling. Together, our results and recent studies argue that the early medulla is organized into common domains comprising processes bound for different layers, and that discrete layers later emerge through successive interactions between processes within domains and developing layers.
Brain function relies on precise networks of synaptic connections. The segregation of these connections into discrete layers allows different aspects of information to be processed in parallel, and is a conserved feature of neural networks in both vertebrates and invertebrates. The medulla neuropil in the Drosophila brain plays a key role in processing visual information. It is analogous in structure and cellular diversity to the inner plexiform layer (IPL) in the vertebrate retina (Sanes and Zipursky, 2010). In both structures, cell bodies remain separate from their axons and dendrites, which form laminated structures within which synaptic connections between specific cells are formed. In each structure processing of multiple locations in visual space occurs in parallel, by discrete units called columns in the medulla and less well defined columnar-like mosaic unit structures in the IPL. As a step towards understanding how such layered structures form during development, we have taken a genetic approach to identifying the mechanisms regulating the targeting of discrete neurons to specific layers of the medulla.
The medulla contains the processes of ~40, 000 neurons. Medulla layers reflect the repetitive distribution of an ensemble of neurons each with a unique morphology. Some axon terminals and dendritic arbors overlap precisely, while others occupy mutually exclusive domains. Using these criteria, Fischbach and Dittrich (K.F. Fischbach, 1989) divided the medulla into 10 layers: the outer layers (M1-M6), the inner layers (M8-M10), and the serpentine layer separating them (M7) (Figure 1A). Although the position of axon terminals and dendritic arbors is largely predictive of synaptic connections between neurons, en passant synapses also form between processes in other layers. In addition to connections formed between elements within a column, connections are made between processes spanning multiple columns, thereby integrating information between different parts of the visual field. There are processes from perhaps 100 different neuronal cell types within each column. The cellular and molecular logic regulating the formation of the medulla circuitry remains poorly understood.
A particularly attractive feature of studying the mechanisms regulating the formation of connections between neurons within the medulla is the availability of reagents for genetically manipulating and visualizing specific neurons during development. Initial studies focused on the mechanisms regulating the projections of photoreceptor neurons R7 and R8 to their specific target layers, M6 and M3, respectively (Figure 1A). Though highly homologous, R7 and R8 neurons project to their target layers using different molecular strategies. R7 target specificity requires the classical Cadherin, N-cadherin (CadN) (Lee et al., 2001), receptor tyrosine phosphatases (Ptp69D and Lar) (Clandinin et al., 2001; Maurel-Zaffran et al., 2001; Newsome et al., 2000), and a novel cell surface molecule (K. Pappu and SLZ unpublished results). By contrast, target layer specificity of R8 does not require these cell surface proteins, but rather relies on the netrin/Frazzled signaling pathway (Timofeev et al., 2012) and a leucine-rich repeat protein called capricious (Shinza-Kameda et al., 2006). In wild type animals R8 growth cones initially target to the surface of the medulla and after a waiting period of several days they extend into the medulla and terminate in their final layer, M3 (Ting et al., 2005). In the absence of Frazzled/Netrin signaling, R8 growth cones remain at the surface of the medulla, whereas in the absence of capricious R8 growth cones exhibit several defects, including projecting past M3 to the M6 layer. Recent studies from Salecker and colleagues support the view that R8 growth cones respond to netrin, selectively secreted by growth cones of lamina L3 neurons. Netrin becomes concentrated within the M3 layer in a manner coincident with the targeting of L3 growth cones to M3. L3 targeting occurs prior to the extension of R8 axons from the surface of the medulla (Figure 1B) and relies, in part, upon CadN (Nern et al., 2008). Thus, different growth cones target to the same layer using different molecular and cellular mechanisms and targeting to specific layers relies upon intercellular signaling between specific processes in the developing neuropil.
Here we set out to understand the mechanism by which L3 growth cones target to the M3 layer. We define a two-step process. Initially, L3 growth cones target to a common domain shared by the growth cones of other neurons destined for a different layer. They then undergo a process of sculpting leading to their segregation into the developing M3 layer. We demonstrate that sculpting requires Sema-1a, and that initial targeting relies on the redundant functions of Sema-1a and CadN in a cell-autonomous fashion. Redundancy between the pathways is established by the complementary expression of Sema-1a, CadN, and PlexA within the developing neuropil. These findings suggest a model in which discrete layers are preceded by common domains comprising processes bound for different layers. Specific non-overlapping layers then emerge progressively through intercellular interactions between processes within the same or adjacent domains.
To uncover the intercellular signaling pathways that control the targeting of L3 growth cones, we screened through RNAi constructs directed to mRNAs encoding all predicted cell surface and secreted proteins in the Drosophila genome (~900) (Kurusu et al., 2008). RNAi transgenes were expressed in L1-L5 neurons using the GAL4/UAS system (Brand and Perrimon, 1993). The GAL4 transgene used was expressed selectively in these neurons; expression commenced just prior to axonogenesis and continued into the adult. Changes in L3 layer specificity were determined in adult flies using a membrane-tethered fluorescent protein expressed selectively in L3 neurons by the LexA/LexOP expression system (Lai and Lee, 2006) (Figure 1C).
RNAi constructs targeted to genes encoding four cell surface proteins, Anaplastic lymphoma kinase (Alk), CadN, sema-1a, and TNF-alpha converting enzyme (Tace), yielded reproducible phenotypes. Alk RNAi resulted in a complete loss of L3 neurons (Figure 1D), consistent with MARCM analysis using a strong loss of function Alk mutation (data not shown). Since Alk likely regulates L3 survival or cell fate and not axonal targeting, we did not investigate its role in lamina development further. RNAi directed towards CadN, sema-1a, and Tace caused L3 axons to mis-target to deeper medulla layers (Figure 1D, Tace not shown). We previously demonstrated that CadN regulates L3 targeting (Nern et al., 2008). As deletion of Tace did not disrupt L3 targeting, the RNAi phenotype is most likely due to knockdown of off-target genes. Therefore, we focused on sema-1a, which encodes a transmembrane Semaphorin protein (Kolodkin et al., 1993). Previously, it was shown that sema-1a is required in photoreceptors R1-R6 for proper topographic distribution within the lamina neuropil (Cafferty et al., 2006). In contrast, sema-1a RNAi does not affect L3 topography within the medulla (i.e. L3 axons are still restricted to the correct column) but rather causes defects in layer-specificity. The penetrance of the sema-1a RNAi phenotype was weak (5-10%) and likely reflects an incomplete knockdown of protein levels in L3 neurons, as the phenotype analyzed in null mutant neurons is much stronger (see below).
sema-1a could act autonomously in L3 neurons or non-autonomously in other lamina neurons to control L3 targeting. To distinguish between these possibilities, we first sought to assess whether Sema-1a was expressed on L3 growth cones. Due to the density of processes within the medulla neuropil and the broad expression of Sema-1a within this region (Figure S1A, see below), it was not possible to address this issue using Sema-1a antibody staining. To visualize Sema-1a expression with single cell resolution, we modified the endogenous locus to conditionally express a tagged protein (i.e. in the presence of FLP recombinase (Struhl and Basler 1993)) (Figure 2A, Figure S1B; also done for CadN Figures S1E-S1G). In the absence of FLP, a transcription termination sequence prevents expression of tagged Sema-1a. When FLP is provided it excises this sequence allowing co-expression of tagged Sema-1a and the LexA transcription factor, which in turn expresses a fluorescent protein to provide uniform labeling of L3 neurons, including the cell body, axons and growth cones. Thus, by controlling FLP expression the localization of endogenous Sema-1a can be observed in L3 neurons or other cells of interest.
The expression of constitutively tagged Sema-1a within the medulla matched that of the endogenous protein (Figure S1A), indicating that the tag does not grossly disrupt localization. Expressing FLP with an enhancer specific for L1-L5 (Millard et al., 2007) yielded a scattered distribution of labeled cells (L1-L5) (Figure 2B). Tagged Sema-1a was highly enriched on L3 growth cones in early and mid pupal stages (Figure 2C, Figure S1C). The staining was punctate, and visible on filopodia (i.e. thin processes extending from the growth cone proper) (Figure 2C). Sema-1a was also expressed on the growth cones of other lamina neurons (Figure 2B, Figure S1D) and, thus, it remained possible that it functioned non-autonomously to regulate L3 targeting.
To directly assess if sema-1a was required autonomously in L3 neurons, we performed mosaic analysis with a repressible cell marker (MARCM) (Lee and Luo, 2001). In these experiments, lamina neurons rendered homozygous for a null mutation in sema-1a were sparsely distributed within the brains of otherwise heterozygous animals, and mutant L3 neurons were specifically labeled in adults (Figure S2A). Mutant L3 axons mis-targeted to the M6 layer within the outer medulla (50%; n=161), or beyond the outer medulla into the inner medulla (14%) (Figure 3A). These phenotypes were rescued by selectively expressing a sema-1a cDNA in mutant L3 neurons (Figure 3A). Thus, the MARCM and protein localization studies indicate that Sema-1a acts within L3 growth cones to prevent mis-targeting to deeper medulla layers.
To determine when Sema-1a is required for L3 targeting, we generated mutant neurons using MARCM and analyzed them during development. At 24h after pupal formation (APF), L3 growth cones are elongated and terminate near the proximal edge of the outer medulla spanning several incipient layers (Figure 1B). Between 24 and 48hrs APF, L3 growth cones undergo a process of sculpting that segregates them into the M3 layer (Figure 1B). This process is highly stereotyped between neurons in different animals. While no defect in targeting was observed at 24h APF, at 48h APF sema-1a mutant L3 growth cones extended inappropriately to the edge of the outer medulla (36%; n=220) or beyond (22%) (Figure 3B), in a manner similar to sema-1a mutant L3 neurons in adult animals (Figure 3A). Thus, Sema-1a is required between 24h and 48h APF during the sculpting of L3 growth cones into the M3 layer. This process is defined by extension within the developing M3 layer and retraction away from the proximal edge of the outer medulla (Figure 1B, 30h APF). Extension within M3 is unaffected in sema-1a mutants (Figure 3C) indicating that Sema-1a is required for retraction from the outer medulla edge. These findings demonstrate that Sema-1a regulates the segregation of L3 growth cones into the developing M3 layer from their initial position spanning multiple developing layers.
Sema-1a has been shown to mediate contact repulsion through binding to the transmembrane protein PlexA (Winberg et al., 1998), acting as a ligand (Sweeney et al., 2007; Yu et al., 1998; Zlatic et al., 2009) or a receptor (Cafferty et al., 2006; Godenschwege et al., 2002; Yu et al., 2010) in a context dependent manner. Since sema-1a is required autonomously in L3 neurons, it is likely to function as a receptor. If PlexA is the Sema-1a ligand then disrupting PlexA function should phenocopy loss of Sema-1a activity in L3 neurons. At 30 hours APF, PlexA protein is strongly expressed in a domain juxtaposing the proximal edge of the outer medulla (Figure 3D), in close proximity to the leading edge of the L3 growth cone prior to and during the retraction process (Figure S2B). This pattern of expression is consistent with PlexA acting as a repulsive ligand for Sema-1a in sculpting L3 growth cones. As reagents are not available to selectively disrupt PlexA in processes within the PlexA-rich region (see below for description of neurons expressing PlexA), we knocked down PlexA mRNA using broadly expressed RNAi constructs. Expression of two different RNAi constructs specific for non-overlapping regions of plexA mRNA abolished PlexA staining (Figure 3D, see below), and resulted in the mis-targeting of L3 axons (Figure 3E). The penetrance and expressivity of this phenotype was similar to that observed with sema-1a null mutations in MARCM experiments (Figure 3A). Thus, Sema-1a/PlexA interactions play an important role in sculpting the L3 terminal, specifically in promoting the retraction of L3 growth cones away from the proximal edge of the outer medulla.
The L3 targeting phenotypes observed with sema-1a mutations and plexA RNAi resembled cell-autonomous mis-targeting induced by removing CadN (Nern et al., 2008) (with a deletion that removes CadN1 and adjacent CadN2 (Prakash et al., 2005), see Figure S2C). In adult animals, CadN mutant L3 axons mis-targeted to the M6 layer (77%; n=150) and less frequently to the inner medulla (4%) (Figure 3A). To assess whether Sema-1a and CadN act within the same pathway or alternatively in parallel to regulate L3 targeting, we compared the penetrance and expressivity of phenotypes resulting from the disruption of either gene alone, or both in combination using null mutants and MARCM. If Sema-1a and CadN act in the same pathway, the double mutant phenotypes should resemble those of single mutants. If they act in parallel, the double mutant phenotypes should be more severe than the single mutants alone. Strikingly, in adult animals 81% of the double mutant L3 neurons terminated within the inner medulla (n=150) (Figure 3A), by contrast to the weak effects on inner medulla mis-targeting due to loss of CadN (4%) or Sema-1a (14%) alone. This supports the idea that Sema-1a and CadN act in parallel to regulate L3 targeting. Moreover, that the penetrance of mis-targeting to the inner medulla in double mutants is more than additive in comparison to single mutants indicates that Sema-1a and CadN act in a partially redundantly fashion.
We previously demonstrated that at 24h APF, before segregation into the developing M3 layer, L3 growth cones mutant for CadN target slightly deeper than normal within the outer medulla (Nern et al., 2008). While endogenous Sema-1a is expressed on L3 growth cones at this time (Figures 2B and C), sema-1a mutant L3 growth cones were indistinguishable from wild type (Figure 3B). To assess whether Sema-1a acts redundantly with CadN at this early stage in L3 targeting, we performed additional MARCM experiments. At 24 hours APF, 83% of double mutant L3 axons mis-targeted (n=230), either to the edge of the outer medulla (33%) or beyond it into the inner medulla (50%) (Figure 3B). Mis-targeting beyond the outer medulla in double mutant L3 neurons was more than additive when compared with single mutants (i.e. no sema-1a single mutant growth cones and 13% of the CadN mutant growth cones extended into the inner medulla at 24h APF (Figure 3B)). Thus, in the absence of Sema-1a function, CadN is sufficient for initial growth cone position in the outer medulla. In the absence of CadN function, the growth cones target slightly deeper, but Sema-1a is largely sufficient to keep them within the outer medulla. In the absence of both pathways, the growth cones initially project through the outer medulla. These findings demonstrate that Sema-1a and CadN act redundantly to restrict L3 growth cones to the outer medulla prior to their segregation into the developing M3 layer. Since mis-targeting beyond the outer medulla becomes more severe as development progresses in double mutants, (compare 24h APF with adult), Sema-1a and CadN may also be involved in maintaining the position of L3 growth cones within the outer medulla.
At 24h APF the growth cones of lamina neuron subclasses (L1-L5) overlap within the outer medulla (Nern et al., 2008) (Figure 4A). Although there is significant overlap between almost all subclasses, the growth cones of different subclasses terminate within either distal (L2, L4) or proximal (L1, L3, L5) regions and display unique morphologies. L3 growth cones and those of L1 and L5 neurons overlap within the proximal outer medulla prior to segregating into different layers. This raised the possibility that L1, L3, and L5 growth cones initially target to the same domain within the outer medulla through the combined functions of Sema-1a and CadN. Similar to L3 axons, MARCM analysis at 24h APF revealed that double mutant L1 (79%; n=108) (Figure 4B) and L5 (22%; n=77) (Figure 4C) axons mis-targeted to the inner medulla. A significant fraction of the double mutant neurons exhibited normal targeting and morphology (i.e. L1, 21%; L5, 78%; also seen in L3 neurons at 24h APF, 17%) indicating that other genes contribute to restricting their growth cones to the outer medulla. In particular, mis-targeting in double mutant L5 neurons was significantly less frequent than in double mutant L1 and L3 neurons. This may reflect a difference in when L5 axons project into the medulla compared with L1 and L3 axons, although the relative timing of their initial projections is not known. The penetrance and expressivity of L1 and L5 double mutant phenotypes increased in adult animals, analogous to L3 neurons, suggesting that Sema-1a and CadN are also required to maintain growth cones within the outer medulla. Indeed, nearly all double mutant L1 axons (95%) mis-projected into the inner medulla or into deeper brain regions (n=173) (Figure 4B), and 100% of double mutant L5 axons mis-targeted into the serpentine layer, with many of these (41%) extending into the inner medulla (n=59) (Figure 4C). As with L3 neurons, in adult animals mis-targeting into the inner medulla in double mutants is more than additive when compared to single mutants (Figures 4C, S3A and B), indicating that Sema-1a and CadN also function redundantly in L1 and L5 neurons. The morphology and behavior (i.e. how far beyond the outer medulla they extend) of sema-1a, CadN double mutant L1, L3, and L5 axons is divergent, and likely reflects intrinsic differences between the neurons. This can also be seen in their unique growth cone morphologies in wild type conditions (Figure 4A). Importantly, a commonality between the phenotypes is mis-targeting beyond the outer medulla. Thus, Sema-1a and CadN act in parallel to restrict the growth cones of L1, L3, and L5 neurons to a common domain within the outer medulla.
The remaining lamina neuron subclasses, L2 and L4, initially project axons to more distal positions within the outer medulla (Nern et al., 2008) (Figure 4A). By contrast to L1, L3 and L5 axons, L2 and L4 axons lacking Sema-1a target normally (Figures S3C and D). Simultaneously disrupting sema-1a and CadN caused morphological defects in these neurons. For example, L2 axons still targeted to the M2 layer, but their terminals displayed abnormal shapes and often had swellings within the M1 layer (Figure S3C). Most of the double mutant L4 axons failed to extend to the M4 layer (Figure S3D). This was previously observed in CadN single mutant L4 neurons and is thought to occur after initial targeting within the outer medulla (i.e. after 24h APF) (Nern et al., 2008). Despite these morphological defects, 94% of L2 axons (n=113) and 87% of L4 axons (n=174) still terminated within the distal outer medulla (Figure S3C and D), demonstrating that Sema-1a and CadN are largely dispensable for this targeting step. The extensive overlap between L2 and L4 growth cones within the proximal outer medulla at 24h APF (Figure 4A) suggests they too may initially target to a common domain, but through an alternative mechanism.
To understand how Sema-1a/PlexA repulsion could combine with CadN mediated homophilic adhesion (Iwai et al., 1997) to control growth cone position within the outer medulla, we examined the expression patterns of Sema-1a, PlexA, and CadN in early pupal stages through antibody staining. At 12h APF, PlexA was concentrated within the nascent serpentine layer, underneath the common domain containing the growth cones of L1, L3 and L5 neurons (Figures 5A and C). The targeting of these growth cones to this common domain in the outer medulla is tightly coordinated with PlexA expression within the serpentine layer. By contrast to PlexA, Sema-1a is expressed on L1, L3 and L5 growth cones (Figures 2C and S1D) and on other processes within the outer and inner medulla (Figures 5A and C). These complementary patterns persisted at 30h APF (Figures 5B and D), when all lamina neuron growth cones have reached the medulla. As PlexA staining overlaps with the fibers of developing medulla tangential neurons (Figures 5E-F, and S4A), it is likely that they are the source of PlexA, although it remains possible that other cells projecting into this fiber bundle produce PlexA. CadN is not expressed within the serpentine layer (Figures 5G-H, and S4B), but is expressed in lamina neurons (Nern et al., 2008) and on other processes within the outer and inner medulla analogous to Sema-1a. Based on this analysis we propose a model that describes how the Sema-1a/PlexA and CadN pathways act in parallel to restrict L1, L3, and L5 growth cones to a common domain within the outer medulla (Fig. 6A). As L1, L3, and L5 growth cones expressing Sema-1a and CadN project into the medulla, they are met by medulla tangential fibers expressing PlexA. Interaction between Sema-1a and PlexA induces repulsive signaling that prevents further extension of the growth cones. At the same time, CadN, not expressed by medulla tangential neurons but highly concentrated within the outer medulla, mediates homophilic adhesion within the outer medulla. This occurs between lamina growth cones, between lamina growth cones and other CadN expressing processes, or both, and acts to restrict L1, L3, and L5 growth cones.
Although the growth cones of L1, L3, and L5 neurons target to different layers, they initially overlap within a common domain in the outer medulla (Figure 4A). Based on biochemical interactions (Iwai et al., 1997; Winberg et al., 1998), and the mis-targeting phenotypes and protein expression patterns described in this paper, we envision that CadN-dependent adhesive interactions restrict processes to the outer medulla and that PlexA-expressing tangential neurons prevent Sema-1a expressing growth cones from projecting into the inner medulla (Figure 6A). L2 and L4 growth cones also appear to initially target to a common domain within the distal outer medulla (Figure 4A), but do not require Sema-1a and CadN for this targeting step (Figure S3C and D), and thus utilize an alternative mechanism. Interestingly, the morphology of L2 and L4 neurons does rely on Sema-1a and CadN function, indicating that within lamina neurons, these molecules regulate different aspects of targeting. This is supported by the expression of Sema-1a and CadN in all lamina neuron subclasses during development (Figures 2B, S1D and G).
Our findings are reminiscent of recent studies in the mouse IPL (Matsuoka et al., 2011), in which Kolodkin and colleagues demonstrated that the processes of different subclasses of PlexA4-expressing amacrine cells are segregated to different OFF layers and that this requires both PlexA4 and Sema6A. Although these proteins act in a more traditional fashion as a receptor and ligand, respectively, they are expressed in a complementary fashion early in development when the developing neuropil is very thin, with PlexA expressed in the nascent OFF layer and Sema6A in the developing ON layers. This raises the intriguing possibility that, as in the medulla, different cells initially target to common domains from which they then segregate into discrete layers. As Cadherin proteins are differentially expressed in a layered fashion in the developing IPL (Honjo et al., 2000) and defects in targeting are incomplete in both Sema6A and PlexA4 mutants (Matsuoka et al., 2011), it is possible that, as in the medulla, Semaphorin/Plexin repulsion acts in parallel with cadherin-based adhesion to control layer-specific patterning within the developing IPL.
Taken together these studies suggest that the restriction of processes to a common domain prior to their segregation into distinct layers may be a developmental strategy used in both the medulla and the vertebrate IPL. This step-wise process may represent a more general strategy for reducing the molecular diversity required to establish synaptic connections by limiting the potential synaptic partners that growth cones and nascent dendritic arbors encounter within the developing neuropil.
After targeting to a common domain within the outer medulla, L3 growth cones undergo stereotyped changes in shape and position that lead to segregation into the M3 layer (Figure 1B). Initially, L3 growth cones are spear-like, spanning much of the depth of the incipient outer medulla. They then expand and elaborate a myriad of filopodia, before resolving into flattened synaptic terminals within the M3 layer. This transformation is marked by two prominent steps (see Figures 1B and and6B):6B): extension of processes from one side of the lateral region of the growth cone into the incipient M3 layer and retraction of the leading edge of the growth cone from the incipient M5 layer (part of the domain shared by L1 and L5 growth cones).
We suggested previously that CadN may regulate the extension within M3, as this step is partially perturbed in CadN mutant growth cones (Nern et al., 2008). However, as CadN mutations affect the initial position of L3 growth cones within the outer medulla (Figure 3B), the extension defect within the M3 layer may be indirect. By contrast, in sema-1a mutant growth cones, initial targeting is indistinguishable from wild type, so defects in retraction away from the incipient M5 layer are likely to reflect a direct role for Sema-1a in this later step in growth cone re-organization. PlexA RNAi phenocopies a sema-1a null mutation and, thus, PlexA is also required for retraction (Figure 3E) and is likely to function on medulla tangential fibers where it is most strongly expressed. In support of this, the tip of the L3 growth cone that retracts is in close proximity to these PlexA-expressing fibers (Figure S2B).
The function of Sema-1a/PlexA signaling in sculpting L3 growth cones appears to be distinct mechanistically from the earlier role it plays in confining the growth cones to a common domain. During initial targeting, PlexA acts as a barrier to L3 growth cones and prevents them from projecting beyond the outer medulla. Thus, at this early step Sema-1a/PlexA interaction provides a stop signal for the leading edge of L3 (uncovered in double mutants with CadN). In the second step, however, Sema-1a/PlexA signaling promotes retraction into the M3 layer. How these diverse outputs of Sema-1a/PlexA signaling arise is unclear. Sema-1a may be coupled to different downstream effectors at each step, modified by association with other receptor subunits or may be modulated by other extracellular signaling pathways.
CadN may also play a role in the retraction of L3 growth cones away from the domain shared with L1 and L5 growth cones. In early pupal stages, disrupting CadN function, while leaving growth cone morphology largely spear-like, causes L3 axons to project deeper within the medulla (Figure 3B). Under these conditions, Sema-1a function is sufficient to prevent the growth cones from extending beyond the outer medulla (Figure 3B). Subsequently, CadN mutant L3 growth cones fail to move away from the outer medulla's proximal edge into the developing M3 layer, and thus remain within the most proximal layer, M6 (Figure 3A). This suggests that CadN, while acting in parallel with Sema-1a to restrict L3 growth cones to the outer medulla initially, may also be required at later stages for movement of the L3 leading edge into the M3 layer. As CadN has been shown previously to regulate neurite outgrowth over cultured astrocytes (Tomaselli et al., 1988), it may be required for L3 growth cones to move along adjacent processes. However, the initial projection of L3 axons into the medulla is not affected by CadN mutations indicating that other components control this process. It also remains possible that the defect in growth cone retraction results indirectly from CadN's earlier role in targeting, as we have suggested this earlier role may account for the defects in growth cone extension within M3.
Disrupting CadN function in different neurons affects targeting in unique ways. For example, L5 axons lacking CadN target to the proper layer, but extend inappropriately within the layer into neighboring columns (Nern et al., 2008). In addition, CadN mutant R7 growth cones display abnormal morphology and, in contrast to mutant L3 growth cones, initially target correctly but retract to a more superficial medulla region (Ting et al., 2005). Collectively, these findings demonstrate that CadN regulates divergent features of growth cone targeting in different contexts. This likely reflects molecular diversity between different growth cones and illustrates the importance of understanding how molecules act in combination to generate target specificity.
Our studies reported here add to previous findings suggesting that column assembly relies on a precisely orchestrated sequence of interactions between different neuronal cell types (Nern et al., 2008; Timofeev et al., 2012). Here we show that as L1, L3 and L5 growth cones expressing Sema-1a enter the medulla they meet the processes of newly arriving tangential fibers expressing PlexA, which acting in parallel with CadN, prevents extension of these growth cones into the inner medulla (Figure 6A). This timing may permit other Sema-1a expressing growth cones to extend into the inner medulla at earlier stages; these growth cones may then use Sema-1a/PlexA signaling for patterning connections in the inner medulla or deeper neuropils of the lobula complex. Subsequent sculpting of the L3 growth cone, mediated by Sema-1a/PlexA and perhaps CadN, leads to its reorganization into an expanded terminal within M3 (Figure 6B). As L3 growth cones become restricted to the M3 layer, netrin, secreted from L3 growth cones, becomes concentrated within the M3 layer and this, in turn, attracts R8 growth cones to the M3 layer (Figure 6C), as recently described by Salecker and colleagues (Timofeev et al., 2012).
Given the extraordinary cellular complexity of the medulla neuropil, with over 100 different neurons forming connections in different medulla layers, and the few mechanistic clues to layer specific targeting that have emerged so far, we envision a complex interplay between different sets of neurons is required to assemble the medulla circuit. The availability of specific markers for many of these neurons, techniques to follow the expression of even widely expressed proteins at the single cell level as we described here, and the ability to genetically manipulate single cells during development provide a robust system for uncovering the molecular logic regulating the layered assembly of axon terminals, dendritic arbors and synaptic connectivity.
In MARCM experiments Dac-FLP (insert on chromosome III) or 27G05-FLP (insert on X chromosome) was used to generate single, isolated lamina neuron clones. The specificity of Dac-FLP (III) for lamina neurons was previously demonstrated using general Gal4 markers (Millard et al., 2007; Nern et al., 2008) (e.g. Tub-Gal4). The specificity of 27G05-FLP was examined with Act-Gal4. Under these conditions only lamina neurons were labeled (not shown).
The positions of lamina neuron growth cones/axon terminals were determined with respect to those of R7 and R8 photoreceptors stained by mAb24B10. From mid-pupal stages (~55h APF) through adulthood R7 and R8 axon terminals mark the M6 and M3 layers, respectively. In earlier pupal stages, R7 and R8 growth cones mark the proximal and distal borders of the outer medulla, respectively. L1 and L2 growth cones and axon terminals were identified based on morphology and relative position within the medulla. These lamina neurons were the predominant subtypes labeled with tubulin-Gal4. When other lamina neurons were visible the fluorescence intensity was much weaker in comparison to L1 and L2. Strongly labeled neurons in adult animals arborizing in the M1 and M5 layers or also into the inner medulla and beyond (mutant) were designated L1 neurons. Strongly labeled neurons that targeted to the M2 layer were designated L2 neurons.
A NotI site was added to the original pBS-KS-attB1-2-GT-SA plasmid (MiMIC(Venken et al.)) downstream of the HindIII in the polylinker. The sema-1a 3′UTR (isoform-PD) including the stop codon and 292 bp of downstream genomic DNA (according to Flybase R5.30) was amplified from a genomic BAC (Venken et al., 2009) (CH322-04A11) using primers containing HindIII and NotI sites and subcloned into this vector. Overlap-extension PCR was used to join two sequences: The first is a fragment a of the pBS-KS-attB1-2-GT-SA plasmid beginning with the NgoVMI site and ending with the splice accepter. The second is the last 1390 bp of coding sequence from the sema-1a cDNA (isoform PD) followed by a XhoI site. This was then subcloned into the NgoVMI/XhoI sites in the above vector containing the 3′UTR. Sequences for the 2A peptide (Tang et al., 2009) and LexAp65 (Pfeiffer et al., 2010) (separated by a KpnI site) were then amplified with primers containing XhoI and HindIII sites and subcloned into these two sites in the above plasmid. Next a fragment encoding a Gly4SerGly4Ser linker followed by a 3xV5 epitope flanked by XhoI and SalI sites was synthesized and subcloned into the above plasmid's XhoI site leaving it a unique site. Finally this site was used to insert either of the following gene synthesized fragments (each preceded by the same GlySer linker used above): (1) an SV40 termination sequence flanked by FRT sites or (2) a single FRT site to generate the control ‘resolved’ cassette.
The following injections, to generate the inducible tag and a constitutively tagged protein, were done: vas-phiC31 Int / sema-1aMI00031 embryos were injected with Sema-1a-3′ORF-FRT-3xV5-2ALexAp65-UTR-pGTNI or Sema-1a-3′ORF-FSF-3xV5-2ALexAp65-UTR-pGTNI. Transgenic animals were identified by the absence of the yellow gene (Figure S1B).
For all developmental analyses, white pre-pupae were collected and incubated for the indicated number of hours at 25° C. Isolated lamina neurons expressing tagged Sema-1a were generated using Dac-FLP 20 (II), which produced a scattered distribution of labeled lamina neurons. Specific lamina neuron subclasses (e.g. L3) were identified based on cell body position in the lamina, growth cone morphology, and growth cone position within the outer medulla. The positions of lamina neuron cell bodies within the lamina are stereotyped such that L5 cell bodies are closest to the lamina plexus (synaptic neuropil) followed by L4 cell bodies. L1 cell bodies are furthest from the plexus and have a characteristic shape simplifying identification. Unlike other lamina neuron growth cones, those of L3 neurons are characteristically elongated, and therefore are readily identifiable. In addition, the positions of lamina neuron growth cones within the outer medulla are stereotyped with L2 and L4 growth cones located more distally, and L1, L3, and L5 growth cones terminating within the proximal outer medulla (Figure 3A). Experimental genotype: w; sema-1a Artificial exon / Dac-FLP 20; LexAop-myrtdTom / TM2. See Figures 2A and S1B for description of the artificial exon. Control genotype (For comparing the expression of tagged Sema-1a with endogenous Sema-1a): w; sema-1a Artificial exon / Cyo; LexAop-myrtdTom / TM6B. This artificial exon contains the resolved FRT sequence (after FLP induced recombination) and lacks a transcriptional termination sequence.
Using a strategy similar to that used for sema-1a, overlap extension PCR was used to insert the last 4680 bp of the CadN ORF (isoform RD) directly downstream of the splice acceptor of the NotI modified pBS-KS-attB1-2-GT-SA plasmid. A linker on the reverse primer used in this PCR allowed for the insertion of four unique restriction sites, AvrII, SphI, BglII, HindIII, between the end of the ORF and the NotI site in the vector. Next, the CadN 3′UTR and 487 bp of downstream genomic DNA was amplified from BAC DNA and added to the above plasmid using the HindIII and NotI sites. A glycine linker with the first FRT site was then added to this plasmid in between the AvrII and SphI sites using annealed oligos. The SV40 termination sequence followed by the second FRT was amplified from the previously used gene synthesized fragment above and inserted into the SphI site. Next, sequences encoding a glycine linker followed by a 3×V5 epitope were also amplified from the above gene-synthesized product and inserted in between the SphI and BglII sites. Finally, sequences encoding the 2A peptide and LexAp65 were amplified and inserted into the BglII, HindIII sites in a manner similar to that done for sema-1a.
The above construct, CadN-3′ORF-FSF-3xV5-2ALexAp65-UTR-pGTNI, was injected into vas-phiC31 Int; CadNMI00393/CyO embryos and transgenics were identified as was done with sema-1a above(Figure S1E). Lines containing a constitutively expressed version of this transgene (Figure S1F) were generated by permanently excising the transcription termination sequence by expressing FLP in the germline.
Exactly as done with sema-1a above.
A collection of RNAi constructs including most of the predicted cell-surface and secreted genes were screened. The RNAi constructs were assembled from the GD and KK libraries of the VDRC, and the Valium libraries from the TRiP collection. The L3 RNAi screen was carried out at 28° C (to increase RNAi expression through elevated Gal4 activity) and in the presence of a UAS-Dcr2 transgene which was previously shown to increase the level of mRNA knockdown (Dietzl et al., 2007).
Experimental genotype: UAS-Dcr2 / w (yv); 22E09-LexA, LexAop-myrtdTom / +; 9B08-Gal4 / UAS-RNAi. 22E09-LexA is specific for L3 neurons in late pupal stages (after 72h APF) and adult animals. 9B08-Gal4 is specifically expressed in lamina neurons from the 3rd Instar larval stage until adulthood. It is also expressed in some medulla neurons and in a subset of photoreceptors early in pupal development. RNAi constructs were screened individually (constructs were localized to the second or third chromosome). Crosses were incubated at 28° C. The brains of adult flies were dissected and stained as described below (Histology). 2 brains per line were dissected and each was imaged independently, in a blind manner (see Imaging). Lines in which both brains gave the same phenotype were retested more carefully in a secondary RNAi screen with additional RNAi constructs. CadN was the only gene besides sema-1a to have a reproducible phenotype.
plexA RNAi was performed at 25° C in the presence of UAS-Dcr2. Eggs were laid on grape plates with yeast extract to prevent overcrowding and provide easy access to food to increase viablility. Two different RNAi constructs (Bloomington 30483, previously reported plexA RNAi construct (Sweeney et al., 2007) from the VDRC collection) specific for non-overlapping sequences within plexA were used in these experiments.
Experimental genotype: w; Act-Gal4 / 22E09-LexA, LexAop-myrtdTOM; plexA RNAi / UAS-Dcr2
L4 neurons were labeled through use of apterous-Gal4, an interruption cassette (containing a transcription termination sequence flanked by FRT sites), and 27G05-FLP.
Experimental genotype: 27G05-FLP; apterous-Gal4, UAS-FRT-STOP-FRT / Cyo, Kruppel-Gal4, UAS-GFP; TM2 / TM6B
Apterous-Gal4 is expressed in many medulla neurons but is specific for L4 neurons within the lamina during development (early pupal stages through adulthood). 27G05-FLP restricts FLP expression to lamina neurons, thus only L4 neurons are labeled by apterous-Gal4.
Primary antibodies. Rabbit pAb α-Sema-1a(Yu et al., 1998) (1:3000); rabbit α-PlexA (Sweeney et al., 2007) (1:500); rat mAb DN-Ex8 (Iwai et al., 1997) (α-CadN, 1:20); chicken pAb α-GFP (abcam; 1:1000); 9E10 α-c-myc (Calbiochem; 1:50); mouse mAb24B10 (Zipursky et al., 1984) (1:20); rabbit pAb α-DsRed (Clontech; 1:200); and mouse mAb α-V5 (Serotec; 1:200). Secondary antibodies: Goat α-rabbit 568 (Molecular Probes; 1:500); goat α-mouse 647 (Molecular Probes; 1:500); goat α-chicken 488 (Molecular Probes; 1:500); and goat α-rat 488 (Molecular Probes). Phalloidin-647 (protein, Molecular Probes; 1:100).
9-9-Gal4 (L3) and 6-60-Gal4 (L5) were previously described (Nern et al., 2008) and generously provided by Ulrike Heberlein. 22E09-LexA (L3), 31C06-Gal4 (L4), 9B08-Gal4 (L1-L5), 65H10-Gal4 (MeTs), and 27G05-FLP (X) were generated at the Janelia Farm Research Campus (Pfeiffer et al., 2010). Apterous-Gal4 and Tubulin-Gal4 are available from the Bloomington stock center. Generation of the Dac-FLP construct was previously described (Millard et al., 2007). Dac-FLP 20 was generated by transposon induced hopping of the original Dac-FLP construct onto the second chromosome.
Histology was performed as described previously (Nern et al., 2008) with minor modification. Fly brains were fixed with PLP (4% paraformaldehyde, 75 mM lysine, 37 mM sodium phosphate buffer pH 7.4) for 25′ at RT. Samples were incubated with primary and secondary antibodies for 2 days each at 4°C. Brains were mounted in Slow Fade Gold anti-Fade Reagent (Molecular Probes).
As previously described (Nern et al., 2008).
We thank Alex Kolodkin and Zhuhao Wu for providing Sema-1a and Plexin A antibodies and a PlexA BAC, Liqun Luo for flies used in genetic mosaic analyses involving sema-1a, Yong Rao for flies used in sema-1a rescue experiments, and Koen Venken for providing reagents for the tagging of endogenous Sema-1a and CadN. For critical reading of the manuscript, we would also like to thank Alex Kolodkin, Kelsey Martin, Sean Millard, Alvaro Sagasti, Kota Saito, and members of the S.L.Z. laboratory. This work was supported by the Jane Coffin Childs Medical Memorial Research Fund, the Human Frontiers Science Program, and the Howard Hughes Medical Institute.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
M.Y.P and W. T contributed equally to this work.