Cell-type Specific Labeling of Synapses in vivo through Synaptic Tagging with Recombination (STaR)

Cell-type Specific Labeling of Presynaptic Sites with the STaR Method

The presynaptic active zones in Drosophila are commonly marked by T-shaped electron dense structures (i.e. T-bars) visualized by Transmission Electron Microscopy (TEM). Similar to the presynaptic ribbons at vertebrate synapses, T-bars are thought to serve as platforms for synaptic vesicle release. They are present at many synapses in Drosophila, including the vast majority, if not all, of the synapses in the visual system, as well as many synapses in the central brain (Butcher et al., 2012; Prokop and Meinertzhagen, 2006; Yasuyama et al., 2003). T-bars are used as the key criterion for synapse identification, as postsynaptic densities are poorly resolved using TEM in flies. T-bars comprise clusters of Bruchpilot (BRP) protein, the single CAST/ERC family member in Drosophila (Fouquet et al., 2009; Kittel et al., 2006; Wagh et al., 2006). By confocal microscopy these clusters appear as fluorescent puncta when stained with BRP antibody, and each punctum correlates with one presynaptic active zone (Fouquet et al., 2009; Hamanaka and Meinertzhagen, 2010). Since BRP protein is broadly expressed in the fly CNS, it is generally not possible to assign BRP puncta to processes of specific cell types, due to the small size of processes and the density of BRP puncta within the neuropil (e.g. see Figure 1B).

To visualize presynaptic sites in specific cell types, we devised a genetic strategy, Synaptic Tagging with Recombination (STaR), to label endogenous BRP protein only in processes of identified cells (Figure 1A). We modified the endogenous brp sequence in a BAC by inserting a V5 tag into the brp coding region downstream from transcriptional and translational stop sequences flanked by FRT recombination sites (Golic and Lindquist, 1989; Southern et al., 1991; Venken et al., 2006; Warming et al., 2005). In the absence of FLP recombinase, the BAC encoded BRP protein is not tagged. By contrast, cell-type specific expression of FLP recombinase induces recombination between the two FRT sites to excises the stop sequence, allowing the expression of V5-tagged-BRP only in cells of interest.

To visualize the morphology of cells expressing tagged BRP, and thus to associate BRP puncta with specific neuronal processes, we inserted a 2A-LexA cassette downstream of the V5 tag; the 2A peptide allows for co-translation of LexA which, in turn, drives expression of, for example, myristoylated-tandem-Tomato (myr-tdTomato) from the LexAoP enhancer (Lai and Lee, 2006; Ryan and Drew, 1994). The modified BAC was introduced into flies as a transgene (Groth et al., 2004). When combined with various cell-type specific FLP recombinases, V5-tagged BRP puncta can be detected in identified neurons marking individual presynaptic sites (Figures 1A, 2, 3 and S3). In order to label presynaptic sites in live animals without relying on immuno-staining, we also prepared a GFP-tagged version of the BRP BAC construct with a similar design (see Experimental Procedures and Figure 8).

The BAC-encoded modified BRP proteins were expressed in the same manner as the unmodified protein from the endogenous locus. In transgenic flies carrying a BAC encoding the constitutively tagged BRP (i.e. brp-FRT-V5-2A-LexA or brp-FRT-GFP), the tagged BRP expression pattern was indistinguishable from BRP expression pattern in the wild type flies both at adult stage and during development (Figure 1B, S1A and S2A). These BACs also rescued lethality completely in brp mutants. In addition, the BRP expression pattern did not change with gene dosage, as it remained the same in flies carrying one to four copies of the brp gene (Figure S1B and S1C).

BRP Puncta in Light Microscopy Match T-bars Visualized by EM

To address whether the BRP marker reliably marks presynaptic sites, we assessed the distribution of BRP puncta in a variety of cell types in the visual system, using various cell-type specific drivers to express FLP recombinase in identified neurons (Figure 2 and S3). As in EM studies, BRP puncta in lamina monopolar neurons L1-L5 were predominantly localized to their axonal terminals in the medulla neuropil (Figure 2B), while L2 and L4, also, elaborated a small number of presynaptic sites in proximal regions of their processes within the lamina neuropil (arrowheads in Figure 2B) (Meinertzhagen and O'Neil, 1991; Rivera-Alba et al., 2011; Takemura et al., 2013; Takemura et al., 2008). In transmedulla neuron Tm9, the vast majority of BRP puncta were observed at their axon terminals in the lobula neuropil (arrows in Figure 2C), while very few BRP puncta were seen in their dendrites in the medulla (arrowheads in Figure 2C). This distribution is consistent with results from EM analysis (Takemura et al., 2013). In addition, we were able to label the presynaptic sites in large interneurons like Dm6 and Dm8 (Figure 2D and S2D). Determining the distribution of synapses in these neurons is difficult at the EM level, due the large volume of neuropil their branches span. In summary, the BRP marker accurately reflects the distribution of presynaptic sites in specific neurons as determined by EM.

We next assessed whether the BRP marker could be used to quantify the number of presynaptic sites in specific neurons. Initially, we compared the number and distribution of BRP puncta with that of T-bars in photoreceptor neurons, as assessed by EM. The fly compound eye contains some 750 ommatidia, each comprising 8 photoreceptor neurons falling into 3 classes: R1-R6, R7 and R8 (Clandinin and Zipursky, 2002). The R1-R6 neurons terminate within the lamina neuropil and are primarily presynaptic to lamina neurons (Figure 3A) (Meinertzhagen and O'Neil, 1991; Rivera-Alba et al., 2011). R7 and R8 axons innervate the medulla neuropil and are primarily presynaptic to specific medulla neurons (Gao et al., 2008; Takemura et al., 2013; Takemura et al., 2008). To assess the number of BRP puncta in R1-R6, we used a pan-photoreceptor FLP (panPR-FLP) to activate the BRP marker in all photoreceptor neurons and assessed the number and distribution of presynaptic sites in the lamina (Figure 3). Two FLP recombinases, expressed selectively in R7 or R8 neurons respectively, were used to examine the presynaptic sites along the length of R7 or R8 axons in the medulla (Figure 3D).

BRP puncta in different classes of photoreceptor neurons were quantified by taking stacks of cross-section images of the axons (right panels in Figure 3B and bottom panels in Figure 3D). BRP puncta were distributed uniformly along the entire length of R1-R6 axons within the lamina neuropil (Figure 3B and C). The number of BRP puncta was remarkably similar to T-bars assessed using EM (Table 1) (Meinertzhagen and O'Neil, 1991; Meinertzhagen and Sorra, 2001; Rivera-Alba et al., 2011). R7 and R8 axons span the outer 6 and outer 3 layers of the medulla neuropil (i.e. layer M1-M6 and layer M1-M3), respectively. As in the EM reconstruction studies (Takemura et al., 2013; Takemura et al., 2008), each R8 axon elaborates presynaptic sites along its length spanning layers M1-M3. By contrast, the presynaptic sites in R7 axons preferentially localize to the M4-M6 layers (Figure 3D and E). As with R1-R6 neurons, the number and distribution of BRP puncta in R7 and R8 neurons was very similar to that of T-bars assessed by EM (Table 1 and Figure 3E) (Takemura et al., 2013; Takemura et al., 2008). Thus, STaR provides an efficient means of quantifying synapses by light microscopy.

Quantification of BRP puncta for several other neuronal classes also correlated well with T-bars analyzed by EM (Table 1) (Meinertzhagen and O'Neil, 1991; Meinertzhagen and Sorra, 2001; Rivera-Alba et al., 2011; Takemura et al., 2008; Takemura et al., 2013). The high correlation between the two methods suggests that the number of presynaptic sites in each cell type is largely stereotyped. In addition, we were able to observe small variations in the number of presynaptic sites among individual neurons of the same class and among individual animals (see Standard Deviations in Table 1 and data not shown). Thus, comparison between STaR and SSEM provides a rapid means of verifying SSEM and conversely the large number of different cells analyzed by STaR permits an assessment of the variability in synapse number not readily achieved by EM.

To assess the importance of using synaptic markers expressed at endogenous levels for mapping synapses, we compared the GFP-tagged BRP STaR marker to the Gal4-driven UAS-BRP-GFP marker in R7 and R8 neurons, respectively (Brand and Perrimon, 1993; Wagh et al., 2006). The protein encoded by UAS-BRP-GFP accumulated in a diffuse pattern within the R7 and R8 axon terminals and thus failed to mark individual presynaptic sites, in sharp contrast to the endogenous STaR marker (Figure 3F and S4). Thus, it is not possible to quantify synapses using a GAL4/UAS version of tagged BRP in these neurons. Similar results were observed with another commonly used presynaptic marker UAS-synaptotagmin-GFP (Figure S4C). Thus, the STaR method, but not GAL4 mediated expression of tagged presynaptic markers, facilitates quantification of synapses within the CNS.

Cell-type Specific Labeling of Postsynaptic Sites with the STaR Method

We sought to generate an endogenous marker that selectively marks postsynaptic sites in specific neurons. In contrast to BRP that marks virtually all presynaptic sites, a complementary marker for postsynaptic sites in flies is not known. As an alternative, we chose to tag neurotransmitter receptors in an inducible fashion to demarcate postsynaptic sites. As a proof-of-principle example, we tagged the histamine-gated chloride channel 2 (Ort), the histamine receptor expressed in the postsynaptic partners of the histaminergic photoreceptor neurons, to demonstrate the feasibility of this approach (Gao et al., 2008; Gengs et al., 2002; Pantazis et al., 2008; Stuart et al., 2007).

We modified the endogenous ort sequence in a genomic construct to encode a different inducible epitope tag named OLLAS (Figure 4A) (Park et al., 2008). The OLLAS tag was inserted immediately downstream of the Ort signal sequence (i.e. at the N-terminus of the mature protein). The modified Ort protein is expressed in the same spatiotemporal fashion and at the same level as the unmodified protein from the endogenous locus (Figure S2B and S5). To achieve cell-type-specific expression, an FRT-flanked stop cassette was inserted into the first intron upstream of the translational start. This cassette included a splice acceptor (SA) upstream of the translational and transcriptional stop sequences in the cassette that prevented expression of the ort gene. When combined with a cell-type specific FLP recombinase, the stop cassette was removed, allowing the OLLAS-tagged Ort protein to be expressed in target neurons marking the postsynaptic sites. As insertion of the 2A-LexA sequence at the 3’ end of the ort open reading frame disrupted Ort protein expression, it was not possible to use this strategy to independently assess the morphology of the cell (see Discussion). Nevertheless, by using highly cell-type specific enhancers to drive the FLP recombinase, we were able to reliably label histamine receptors in identified neurons (see Figure 4).

Cell-type specific FLPs were used to selectively activate the Ort marker in L3 neurons (postsynaptic to R1-R6 neurons) and Dm8 neurons (postsynaptic to R7 neurons) (Figure 4B-G) (Gao et al., 2008; Rivera-Alba et al., 2011). L3 neurons extend multiple primary dendrites from the axon shaft with short spike-like secondary dendrites extending between neighboring R1-R6 axons, as shown by the targeted expression of the dendritic marker Denmark via the Gal4-UAS system (Figure 4B and C) (Brand and Perrimon, 1993; Nicolai et al., 2010). OLLAS-tagged Ort in L3 preferentially localized to secondary dendrites (arrowheads in Figure 4D), although some signal was observed along the primary dendrites and the proximal region of the axon shaft (Figure 4D). When we compared the expression pattern of the Ort marker to a general membrane marker (i.e. myr-GFP) expressed in the same L3 neurons, a clear difference in intensity at specific subcellular locations was observed, supporting that the preferential localization of OLLAS-Ort to L3 secondary dendrites is not due to the overall lower expression level of the endogenous construct (Figure S5B).

Dm8 neurons extend a single axon to the M6 layer and arborize extensively within this layer (Figure 4E and E’) (Gao et al., 2008). OLLAS-labeled puncta decorated Dm8 processes and overlapped extensively with the presynaptic R7 terminals (Figure 4F-G’). Taken together, these results demonstrate that the Ort marker is selectively enriched in regions in close proximity to presynaptic terminals of photoreceptor axons.

Co-labeling Presynaptic and Postsynaptic Sites with STaR using Two Different Recombination Systems

In order to simultaneously label presynaptic and postsynaptic sites in partner neurons, we incorporated two non-overlapping recombination systems in the same animal (Figure 5) (Nern et al., 2011). The major output synapses of R1-R6 photoreceptor neurons are tetrads (i.e. one presynaptic site opposing four postsynaptic elements); L1 and L2 lamina neurons are postsynaptic at all tetrads and L3 is only postsynaptic at a subset of these with the remainder of the postsynaptic elements provided by amacrine cells or glia (Figure 5A) (Meinertzhagen and O'Neil, 1991; Rivera-Alba et al., 2011). Endogenous Ort is expressed in lamina neurons L1- L3, but neither in L4 nor L5 (Gao et al., 2008). To tag Ort in L1-L3, FLP recombinase was expressed in all lamina neurons. To co-label the presynaptic sites we generated a new version of the BRP marker by replacing the FRT sites with R recombinase recognition sites (RSRT) (Nern et al., 2011). R recombinase was then expressed in all photoreceptor neurons to induce V5 labeling of BRP in R1-R6 neurons (Figure 5A). Combining the two labeling systems in the same animal resulted in matching of V5 puncta with concentrated OLLAS staining in synaptic partners (Figure 5B). This matching was not observed when we co-expressed the BRP marker in R1-R6 neurons and a general membrane marker (myr-tdTomato) in L1-L3 neurons (Figure S6).

Using a similar scheme, we labeled V5-tagged BRP in the presynaptic R7 neurons and OLLAS-tagged Ort in the postsynaptic Dm8 neurons in the medulla (Figure 5C). The juxtaposition of V5 puncta and concentrated OLLAS staining shows that STaR allows co-labeling of presynaptic and postsynaptic sites simultaneously in synaptic partners in a complex neuropil.

Formation of Presynaptic Sites in Developing Neurons

The extraordinary density of axonal and dendritic processes of many different neuronal cell types in the developing neuropil hinders the study of synapse formation in the CNS. Here, we took advantage of cell-type specific labeling of presynaptic and postsynaptic sites during development to follow the time course of synapse formation in all three classes of photoreceptor neurons.

During synapse formation, synaptic proteins are trafficked to specific locations and assembled into presynaptic structures (Owald and Sigrist, 2009). As the tagged BRP protein is expressed via its endogenous regulatory mechanisms, the appearance of BRP puncta provides a marker for active zone assembly (Fouquet et al., 2009). We analyzed presynaptic development in photoreceptor neurons by assessing the spatiotemporal accumulation of BRP puncta within them at different times during development.

R1-R6 neurons extend axons into the lamina during the 3rd instar larval and early pupal stages with their growth cones terminating in the developing lamina (Figure 6A). These growth cones rearrange within the lamina between 30 and 40 hours after puparium formation (hAPF) forming nascent lamina cartridges comprising the growth cones of six R1-R6 axons from six different ommatidia. These axons then extend some 20 microns deeper forming the full depth of the lamina neuropil (Clandinin and Zipursky, 2002). Selective tagging of BRP in photoreceptor neurons revealed a few BRP puncta within R1-R6 growth cones at 40hAPF, just prior to extension (Figure 6B and C). By 45hAPF, R1-R6 axons extend an additional 5 microns as they become morphologically transformed from motile growth cones with filopodia to cylindrically shaped axon terminals. As these axons extend, BRP puncta appear at the borders between neighboring axons suggesting that initial presynaptic sites have formed (arrowheads in Figure 6C). Although terminal extension is complete by 65 hAPF, BRP puncta continue to accumulate through the remaining 30 hours of pupal development (Figure 6B, C, H and I).

A similar time course of BRP accumulation was observed in R7 and R8 terminals. Prior to 40hAPF, R7 and R8 growth cones extend into the developing optic lobes in birth order and terminate in intermediate targets (Figure 6D). R8 growth cones reside at the distal edge of the medulla and R7 growth cones occupy a characteristic position proximal to R8 growth cones. At 40hAPF each R8 growth cone harbors 1-3 bright BRP puncta roughly in the center of the growth cone; a small number of BRP puncta were also seen in R7 growth cones (Figure 6E, F and H). Commencing just after 40hAPF, R7 and R8 growth cones extend into deeper layers of the medulla in a synchronous fashion (Nern et al., 2005; Timofeev et al., 2012; Ting et al., 2005). Coincident with this synchronous extension, BRP puncta begin to accumulate in the newly extended R7 and R8 axon terminals also in a synchronous fashion (Figure 6E, F, H and I). This process is very similar to the presynaptic development in R1-R6 neurons in the lamina.

Together, despite differences in their targeting regions and synaptic partners, the time course of presynaptic development for all three classes of photoreceptor neurons is highly similar: the formation of presynaptic sites initiates in the same time window between 40h-53hAPF and is coincident with the extension of the axons marking the transition from growth cones to synaptic terminals. Importantly, when the UAS-BRP-GFP marker was expressed in R7 neurons via Gal4-mediated expression at these developmental stages, strong GFP signal was observed throughout the terminals at both 40hAPF and 53hAPF, failing to reveal the transition of these growth cones to synaptic terminals (Figure 6G and S4). Thus, STaR facilitates analysis of synaptic development at the level of single identified cells in the context of a complex neuropil.

Accumulation of Neurotransmitter Receptors at Postsynaptic Sites

We next followed postsynaptic development of targets of photoreceptor neurons. For this, we focused on L3 neurons in the lamina using targeted expression of Denmark to label L3 dendrites and accumulation of OLLAS-tagged Ort through STaR to label postsynaptic sites. At 24hAPF, filopodia-like dendritic structures extend from the proximal region of the L3 axon (Figure 7A). Between 40-45hAPF, these processes extend into the center of the lamina cartridge between the inner cluster of L1 and L2 axons and the surrounding rosette of R1-R6 axons (Figure 7A and data not shown). By this stage of development, a significant number of presynaptic sites in the R1-R6 axons have formed (Figure 6B, C and H). By 53hAPF, short spike-like secondary dendrites extend out from the primary dendrites and intercalate between the R1-R6 axons. Curiously, these secondary dendrites continue to elongate and surround individual photoreceptor axons at 65hAPF, but later retract (arrowheads in Figure 7A). By 77hAPF, the retraction is complete and the dendrites look morphologically indistinguishable from those of adult L3 neurons. OLLAS-tagged Ort begins to accumulate in L3 dendrites at 77hAPF (Figure 7B), after L3 dendrites have adopted their mature morphology and long after the onset of BRP puncta accumulation in the presynaptic photoreceptor axons.

The late localization of Ort to dendrites, after the appearance of BRP puncta also occurred in the other postsynaptic partners of R1-R6 neurons including L1, L2 and Am cells, as well as in the Ort-expressing cells in the medulla postsynaptic to R7 and R8 terminals (Figure S2B). These findings suggest a common timeline for synapse formation in all three classes of photoreceptor neurons (Figure 7C). Presynaptic sites start to form in the photoreceptor axons in close temporal association with dendritogenesis in target neurons. Dendritic processes of postsynaptic neurons go through dynamic morphological changes. It is only after they adopt their mature morphology, some 30 hours after the onset of dendritogenesis, that neurotransmitter receptors become localized to these postsynaptic sites.

Following the Formation of Presynaptic Sites in Live Animals

The ability to image synapses in real-time in live animals is crucial to reveal dynamic changes during synapse formation as well as changes in synaptic organization as a consequence of activity. To test the utility of the STaR strategy in visualizing synapse formation in live animals, we used an R8-specific FLP recombinase to selectively tag BRP with GFP in R8 neurons and followed the dynamics of this marker during the time window of initial synapse formation (i.e. 42-55hAPF) in live, developing pupae with two-photon microscopy (O.A. and S.L.Z., manuscript in preparation). Gal4-mediated expression of UAS-myr-tdTomato was used in parallel to mark R8 axons. After the pupal case surrounding the head was gently removed, each pupa was mounted on a slide with one eye contacting the coverslip. The objective lens was placed against the eye under the coverslip and the dynamics of the developing R8 terminals were visualized directly through the retina in the underlying optic lobe. This is a non-invasive procedure; fertile flies emerge after pupal development with no morphological defects in the visual system.

Consistent with our observation with the fixed samples, after targeting to the distal edge of the medulla, R8 growth cones remain there until 40hAPF with weak GFP signal localized to the center of the growth cone (Figure 8). After 40hAPF, each R8 growth cone starts targeting to the M3 layer first with the rapid extension of a very thin process (Figure 8, red channel, 42-45hAPF). Once this leading process reaches the M3 target layer, it thickens from the top down and the R8 growth cone begins its transformation into a mature axon terminal with the appearance of many BRP puncta along its length (Figure 8B, 45-52hAPF). Live imaging revealed that stable BRP puncta are largely absent from the leading thin process but rapidly populate the thickening terminal. When individual puncta were tracked between successive time points, we found that stable puncta appear from the proximal to distal locations along the length of the extending terminal (data not shown). In summary, STaR allows imaging of synapse formation in specific neurons in real-time within the CNS non-invasively in intact animals.

Drosophila stocks

Specific fly genotypes in each experiment are described in supplementary methods. Flies were reared at 25° on standard cornmeal/molasses food. Pupal staging was performed by counting the number of hours at 25° after selecting pre-pupa (i.e. 0 hAPF). For instance, 24hrs after the pre-white pupal stage at 25° is 24hAPF.

In addition to the STaR markers presented in this study, we used the following lines: (1) 9-9 Gal4 (L3); (2) Dac-FLP 20 (Pecot et al., 2013); (3) MH56 Gal4 (L3) (Timofeev et al., 2012); (4) Rh4-Gal4 (R7); (5) Rh6-Gal4 (R8) (Tahayato et al., 2003); (6) GMR-FLP (PanPR-FLP) (Pignoni et al., 1997); (7) UAS-FLP (Duffy et al., 1998; Nern et al., 2011); (8) GMR-Gal4 (PanPR-Gal4) (Wernet et al., 2003); (9) UAS-Denmark (Nicolai et al., 2010); (10) UAS-BRP-GFP (Wagh et al., 2006); (11) 20C11(R7)-FLP; (12) Senseless (R8)-FLP; (13) LexAoP-myr-tdTomato in su(Hw)attP2 and su(Hw)attP5; (14) GMR>stop>Gal4; (15) UAS>stop>myr-tdTomato in su(Hw)attP2; (16) UAS-R in attP2 (Nern et al., 2011); (17) UAS-myr-GFP in attP40 (18) UAS-myr-tdTomato in su(Hw)attP2 (Pfeiffer et al., 2010); (19) R27G05-FLP2::PEST in attp40 (PanLN-FLP); (20) R20C11-Gal4 (R7); (21) R24C08-Gal4 (Tm9); (22) R53C12-Gal4 (Mi1); (23) R20G06-Gal4 (Dm6); (24) w; R20C11-AD; R25B02-DBD (C2); (25) w; R26H02-AD; R29G11-DBD (C3); and (26) w; R55F02-AD; R69A01-DBD (Dm8) (Pfeiffer et al., 2008; Tuthill et al., 2013). Stock (4)-(10) were acquired from Bloomington Drosophila Stock Center. Stock (11)-(15) were generated in this study. Stocks (24)-(26) are split GAL4 stocks.

Histology

Histology was performed as described previously with minor modification (Pecot et al., 2013). Fly brains were fixed with PBL (4% paraformaldehyde, 75 mM lysine, 37 mM sodium phosphate buffer pH 7.4) for 25min at room temperature. After multiple rinses in PBS with 0.5% Triton X-100 (PBT), brains were blocked in 10% normal goat serum in PBT (blocking solution) for 1hr. Brains were incubated with primary and secondary antibodies for 2 days each at 4° with multiple rinses in blocking solution in between and afterwards. Brains were mounted in Slow Fade Gold anti-Fade Reagent (Life technologies).

See Supplementary Experimental Procedures for a complete list of primary and secondary antibodies used in this study.

Microscopy and image analysis

Confocal images were acquired with Zeiss LSM780 confocal microscope. The staining patterns were reproducible between samples but overall fluorescence signal and noise unavoidably shows some variation between sections and samples. Some adjustments of laser power, gain and black level settings were therefore made to obtain similar overall fluorescence signals. Single plane or maximum intensity projection confocal images were exported into TIF files using LSM Image Browser (ZEISS).

For quantification of the BRP-V5 puncta, confocal stack images were taken from the optimal orientation (cross-section of the axon axes) to cover the entire axon length. For each cell type, several Z-step values were tested out to find an optimal value (normally between 0.8-1.5μm) to avoid imaging the same puncta redundantly in adjacent sections or missing puncta. The number of BRP-V5 puncta were then scored by two counters (counter #1 and #2 in Table 1) independently by going through the stack images section by section with LSM Image Browser. Counter #2 was an undergraduate student who was unfamiliar with synapses in the fly visual system. Five optic lobes from five animals were imaged and quantified for each cell type. For R1-R6 neurons, three cartridges in each optic lobe were scored, giving a total of 15 cartridges or R1-R6 sets. For L2 neurons, five cells in each optic lobe were scored, giving a total of 25 cells. For the other cell types, six cells in each optic lobe were scored, giving a total of 30 cells for each cell type.

Live imaging of Drosophila pupae

Live Drosophila pupae were staged and mounted on slides at around 36hAPF with one eye contacting the coverslip. Stack images of the R8 axons were taken directly through the retina into the underlying optic lobe every 15 minutes for the next 24hrs with custom 2-photon microscope. These stack images were then reconstructed and aligned to generate maximum-intensity projection images shown in Figure 8. (O.A. and S.L.Z., manuscript in preparation). Detailed information regarding the 2-photon microscope, the imaging set-up and data processing are available upon request.