DIP-γ interacts with four Dpr partners with similar affinities: Dprs 11, 15, 16 and 17 (Figure 1D). A subset of R7 PRs in pupal retina express GFP from the dpr11^MiMIC reporter line (henceforth denoted as dpr11^Mi>GFP). These were confirmed to be yR7 by co-labeling with a Rhodopsin marker, Rh4-LacZ, in late pupae (Figure 1B, E–F). Reporters for Dprs 15, 16, and 17 are not detectably expressed in the pupal retina.

Here, we classify two subtypes of Dm8 neurons, the major synaptic partners of R7 PRs: those which express DIP-γ (yDm8) and those which do not (pDm8) (Figure 1G–G’). We determined the yDm8 population in the adult medullary cortex as those cells that express RFP under the control of a late pupal pan-Dm8 driver, R24F06-GAL4 (Nern et al., 2015) and GFP from the DIP-γ^MiMIC reporter line (henceforth denoted as DIP-γ^Mi>GFP). The pDm8 population was identified as those cells that express RFP but not GFP; there are no known markers or drivers that selectively label pDm8 neurons. We additionally used the transcription factor Dachshund (Dac), which is expressed in both yDm8 and pDm8 neurons (Hasegawa et al., 2011), together with DIP-γ^Mi>GFP to selectively identify yDm8 cell bodies in early pupal stages and to confirm yDm8 identity in adults independently of the driver. yR7 and pR7 ommatidia are present at a ~65y:35p ratio, and because of retinotopy we expect to find a similar ratio of y and p columns in the medulla. We find that yDm8 and pDm8 neurons in wild-type are present at a ratio of 60y:40p, which is comparable to the ratio of the input R7 PRs (Figure 1B,H–I) (Viets et al., 2016).

Dm8 neurons are both unicolumnar and multicolumnar in their coverage of columns in the neuropil. The arbor of each Dm8 contacts 12–16 medulla columns, but most synapses are made with the R7 in the central (home) column (Fischbach and Dittrich, 1989; Gao et al., 2008; Takemura et al., 2013; Takemura et al., 2015). Figure 2A shows a horizontal view (side-view as in schematic in Figure 1C) of a rendering of an R7 terminal and a Dm8 arbor from an EM reconstruction (Takemura et al., 2015). The thick bundle of dendritic processes at the center of the arbor makes extensive contacts with the R7 home column, and contains the majority of R7-Dm8 synapses (Figures 2A and 3). These central dendritic projections extend distally from M6 to M4, and we have denoted them as the ‘sprig’ of a Dm8 arbor. The lateral arbor of each Dm8 overlaps extensively with other Dm8 arbors, but the center of highest arbor density, the sprig, tiles the medulla. The coverage pattern of a typical Dm8 is thus indicative of approximately one cell per column.

Figure 2 with 1 supplement see all

Download asset Open asset

Figure 3

Download asset Open asset

Figure 3—source data 1

This is a table listing non-home column R7 synapses, and R7/Dm8 synapses with Tm5c and Dm9, in addition to the information in Figure 3.

Column two indicates the identities of the input synapses from R7 onto each Dm8. The reconstructed volume is unlikely to cover the entire arbor of these Dm8 neurons, since each Dm8 can contact 12–16 R7 (Gao et al., 2008). However, column two shows that within the reconstructed columns there is no apparent specificity in synaptic connectivity between y and p R7 and Dm8 neurons outside of their home columns. For example, pDm8-B/home receives synapses from four non-home column R7, at least 2 of which are yR7, while yDm8-A receives synapses from two non-home column pR7 and a yR7. Even the Dm8 at the center of the reconstruction (Dm8-B/home) only receives synapses from four non-home column R7 PRs, so many Dm8 contacts with R7 columns may not be associated with input synapses from R7. The total number of non-home column R7 PRs that synapse onto the Dm8 is indicated in the parentheses in column 2. Synapses labeled as ? in column two are from R7 PRs in columns that cannot be assigned as y or p. Also note that (Karuppudurai et al., 2014) found that Tm5a dendrites specifically associate with yR7 axons, but they stated that Tm5b dendrites have no specificity for association with y vs. p R7 axons. Our analysis of the EM reconstruction, however, shows that all 3 pR7 PRs synapse only onto Tm5b and not Tm5a. The conclusion of Karuppudurai et al. (2014) likely arises from the fact that Tm5b usually has two distal dendritic projections. One of these always arborizes with a pR7 axon and receives the pR7 input. The other one can arborize with either a y or a p R7, but does not receive yR7 input, at least not for this set of columns. Four of the 7 Dm8, and 3 of the 8 R7 PRs, also have synapses onto Tm5c neurons. There is no specificity for y vs. p in the connections to Tm5c. Both R7 classes have input and output synapses with Dm9, and both classes of Dm8 have output synapses onto Dm9.

https://cdn.elifesciences.org/articles/48935/elife-48935-fig3-data1-v2.xlsx

Download elife-48935-fig3-data1-v2.xlsx

Although the ratios of the input PRs and output targets match, the Dm8 cell numbers we obtained (266 yDm8 and 179 pDm8) are insufficient for innervation of the expected 730–785 yR7 plus pR7 columns in the medulla (Posnien et al., 2012) even when we account for the fact that some Dm8 neurons (~6%; see below) have two home columns. This suggests that the Dm8 driver is not completely penetrant. Indeed, when we use Dac and DIP-γ^Mi>GFP to determine yDm8 populations, we obtain a larger number (~320), although this is still less than the expected total number of yDm8 neurons, which is 440–480.

To evaluate the specificity of connections between Dm8 and R7 subtypes, we generated single-cell Dm8 (flipout) clones. For yDm8 flipouts, we used an intersectional strategy, since DIP-γ is expressed in many non-Dm8 cells. This employed a split-GAL4 driver that includes DIP-γ^MiMIC GAL4-DBD and R24F06 p65-AD hemi-drivers to selectively label DIP-γ-expressing yDm8 neurons (Figure 2B,D, Figure 2—figure supplement 1C–C’). For Dm8 that do not express DIP-γ (pDm8) flipouts, we used the pan-Dm8 driver and Rh4-lacZ reporter to identify clones whose home column R7 is labeled by the pan-PR marker Chp but lacks LacZ labeling, since there is no pDm8-specific driver (Figure 2C,E). yDm8 and pDm8 arbors have similar morphologies, with the prominent dendritic sprig identifying the home column (Figure 2A–C) (Gao et al., 2008; Nern et al., 2015). Using the yDm8 split-GAL4 driver, we exclusively labeled Dm8 neurons that have yR7 as the home column (43/43 flipouts; Figure 2B,D; Figure 2—figure supplement 1A). This indicates that yDm8 specifically innervate yR7 and avoid pR7 in the home column. We also observed yDm8 with two home columns at a frequency of ~6% (4/64 clones) (Figure 2—figure supplement 1A). All four had both sprigs on yR7 columns (see Figure 5H for an example). Since we find that all DIP-γ-positive Dm8 neurons have yR7 home columns, we can infer that cells labeled using the pan-Dm8 driver that have pR7 home columns must be the DIP-γ-negative pDm8 subtype. Single-cell clones of pDm8 were found to innervate only pR7 (21 pDm8 clones were analyzed; one had two pR7 home columns) (Figure 2C,E, Figure 2—figure supplement 1A). Specificity was further confirmed by labeling mid-pupal optic lobes with DIP-γ antibody and dpr11^Mi>GFP. DIP-γ labeling is seen in the M6 layer only under yR7 that co-express Dpr11 and Chp, and not under pR7 that are labeled with Chp alone (Figure 2F).

We analyzed neighboring R7 contacted by yDm8 and pDm8 in single-cell clones. Outside of their home columns, yDm8 and pDm8 dendritic arbors contact both types of R7. Cross-sectional (top-down) views show that although there is no significant difference in the total number of R7 terminals contacted by the two Dm8 subtypes, there is a bias in the type of R7 they contact (Figure 2D–E,I–J). yDm8 contacts with yR7 columns were significantly higher (~15%) than pDm8 connections with yR7.

The two subtypes of Dm8 are distinguished by the expression of DIP-γ. To evaluate whether the yDm8 and pDm8 populations arise independently or originate as one population, with DIP-γ being switched off in pDm8 neurons at a later time, we used DIP-γ GAL4 to express FLP in a line with a flipout cassette (LexAop-FRT-stop-FRT) driving a GFP reporter and a pan-Dm8 LexA driver and examined GFP labeling in the M6 layer (Figure 2G). If yDm8 and pDm8 originate as separate populations, we would expect gaps to be present in M6 (like those seen with DIP-γ^Mi>GFP (Figure 2H’; schematic in Figure 2—figure supplement 1B) and not a continuous line in M6 (like that seen when both yDm8 and pDm8 are labeled with a pan-Dm8 driver (Figure 2H”)). This is because the conditional DIP-γ GFP reporter would be expressed only in yDm8 and absent in pDm8. The gaps represent ‘pale’ columns with a pR7 and have a pDm8 arbor, which does not label with DIP-γ^Mi>GFP (or with DIP-γ antibody; Figure 2F). Figure 2G shows that there are gaps in the M6 layer, indicating that the two Dm8 subtypes are in fact separate populations.

Most R7 output synapses are made onto Dm8, Dm9 (a multicolumnar medulla intrinsic neuron), Tm5a, Tm5b, and Tm5c (Gao et al., 2008; Karuppudurai et al., 2014; Takemura et al., 2013; Takemura et al., 2015). Tm5a/b/c neurons project to the lobula, and are likely to be the main output neurons of R7 circuits. To define the synaptic connections of these neurons and determine whether they differ between y and p columns, we examined the EM reconstruction of the medulla (Takemura et al., 2013; Takemura et al., 2015). This reconstruction attempted to identify all synaptic connections between the ‘home’ column in the center and the six columns surrounding it (columns A-F in Figure 3 diagram). The surrounding columns (G-R) were also partially annotated, but these columns vary in the extent to which their synapses can be assigned to specific cells. There are eight columns for which R7, Dm8, and Tm5a/b synapses are annotated (home, A, B, D, E, F, J, and K), and analysis of these columns allowed us to derive the conclusions described below.

The morphologies of y and p R7 and Dm8 neurons, as visualized in the EM reconstruction, do not allow us to distinguish their subtypes. However, the main dendritic branch of Tm5a was found to associate with yR7 but not with pR7 axons in a light-level analysis (Karuppudurai et al., 2014), suggesting that yR7 selectively synapses onto Tm5a. Using this hypothesis as a guide, we identified five candidate yR7 PRs (in columns A, E, F, J, and K) and three pR7 PRs (in columns Home, B and D) (Figure 3; y/p column identity indicated in diagram). There are seven Dm8 neurons whose home columns correspond to these eight R7 PRs (Dm8-B/home has two home columns). Each R7 makes many more output synapses onto its home column Dm8 than onto other Dm8 neurons, allowing us to identify B/home and D as pDm8, while A, E, F, J, and K are yDm8 (Figure 3).

To analyze projection neuron input specificity, we counted synapses made by y and p R7 PRs onto Tm5a and Tm5b. The five yR7 make output synapses only onto Tm5a and not Tm5b, which was the basis for their assignment as y (Karuppudurai et al., 2014). The three pR7 synapse onto Tm5b and not Tm5a (Figure 3). Thus, Tm5a and Tm5b represent separate output channels for y (Rh4) and p (Rh3) inputs for this set of columns (see Figure 3—source data 1 legend). Tm5a expresses DIP-γ, while Tm5b does not (Cosmanescu et al., 2018). This suggests that the selective connection of yR7 to both yDm8 and Tm5a in the home column might involve Dpr11-DIP-γ interactions. There are no GAL4 drivers that distinguish between Tm5a and Tm5b, so we cannot determine if yR7-Tm5a connections are altered in mutants.

Dm8 arbors contain both pre- and postsynaptic elements, and Dm8 is presynaptic to Tm5a/b and Dm9. We find that Dm8 output synapses onto Tm5a/b have a similar specificity to the R7 output synapses. Both pDm8 neurons almost exclusively synapse onto Tm5b and not Tm5a, and four of the five yDm8 neurons (E, F, J and K) make more synapses onto Tm5a than Tm5b (yDm8-A, however, makes 10 synapses onto Tm5b-home and only five onto Tm5a-A; Figure 3). Each Dm9 receives input from both types of R7 and Dm8, and makes output synapses onto multiple R7 neurons (Figure 3—source data 1).

R7 synapses are polyadic, and Dm8 and Tm5a/b often sit together at synaptic sites. Figure 4D and E show R7 T-bars (active zone elements marking output synapses) adjacent to both yDm8 and Tm5a, and pDm8 and Tm5b, respectively. In column E, the yR7-E axon terminal, yDm8-E sprig, and Tm5a-E dendritic branch are tightly wrapped around each other (Figure 4A,B). yR7-E T-bars are distributed in layers M4-M6, and are apposed to postsynapses in the yDm8-E sprig and the distal dendritic branch of the Tm5a-E neuron (Figure 4B). Most yDm8-E output synapses are onto Tm5a-E and Dm9 (Figure 3 and Figure 3—source data 1), and these are distributed between the sprig in M4-M5 and the main arbor in M6 (Figure 4B and associated Figure 4—videos 1 and 2). The only pDm8 with one home column in the reconstructed volume is pDm8-D. Although pDm8 usually have robust sprigs (e.g. Figure 2C), pDm8-D has a very thin sprig, and most pR7-D and pDm8-D T-bars are in M6 (Figure 4A,C and associated Figure 4—videos 3 and 4).

Figure 4 with 7 supplements see all

Download asset Open asset

Columns B and home form a two-home column pale circuit (Figure 3). The cells in this circuit are pR7-B, pR7-home, pDm8-B/home (a two-home column pDm8), Tm5b-B, and Tm5b-home (Figure 4—figure supplement 1). The B sprig of pDm8-B/home and one of the dendritic branches of Tm5b-B are both wrapped around the pR7-B terminal, while the pR7-home terminal is more loosely associated with the home sprig of pDm8-B/home and dendritic branches of both Tm5b neurons (Figure 4—figure supplement 1, and Figure 4-associated Figure 4—videos 5 and 6). All pR7-B→Tm5b synapses (14) and most pDm8-B/home→Tm5b synapses (30 vs. 4) are onto Tm5b-B. pR7-home makes similar numbers of synapses onto Tm5b-B and Tm5b-home (Figure 3).

If the yellow circuit is constructed using DIP-γ-Dpr11 interactions, as suggested by our analysis of the EM reconstruction, one might expect that there would be abnormalities in yDm8 arbors in DIP-γ and dpr11 mutants. To assess yDm8 arbor morphology in whole-animal mutants, we generated flipouts in wild-type and mutants and examined arbor morphology (Figure 5A–F; Figure 5—source data 1). The sprig, which is located in the center of the arbor, usually has an expanded region at the distal end in M4/M5 (Figure 5A–B; see also Figure 2A–C). There is a large variation in sprig diameter and height in both wild-type and mutants. Maximum sprig diameter was significantly decreased in both DIP-γ and dpr11 mutants compared to control, but the height of the sprig was relatively unaffected (Figure 5C–F,I,K). The numbers of two-home column single cell clones (with two sprigs) were higher in DIP-γ mutants as compared to controls (Figure 5G–H). We also examined contacts made by the yDm8 arbor to neighboring columns (Figure 5J,L). There was no difference in the percentage of yR7 columns contacted in either mutant, but yDm8 neurons in the DIP-γ mutant displayed an increase over control in the total number of columns contacted.

Figure 5

Download asset Open asset

Figure 5—source data 1

Frequency of two-home column yDm8 clones is increased in DIP-γ mutants.

The percentage of two-home column yDm8 clones in DIP-γ ^-/- mutants is 4.5-fold higher than in wild-type. Quantitation of flipout clones in wild-type, DIP-γ and dpr11 mutants shown. Flipouts were generated with either pan-Dm8 driver R24F06-Gal4 or the yDm8 split-Gal4 driver R24F06-p65.AD; DIP-γ Gal4-DBD. Complete genotypes in Table 3 of Materials and methods.

https://cdn.elifesciences.org/articles/48935/elife-48935-fig5-data1-v2.xlsx

Download elife-48935-fig5-data1-v2.xlsx

To obtain high-resolution views of interactions between yDm8 sprigs and R7 terminals, we examined single cell yDm8 clones in wild-type and DIP-γ mutant brains using expansion microscopy (ExM) (Figure 6 and associated Figure 6—videos 1–4) (Mosca et al., 2017; reviewed by Karagiannis and Boyden, 2018). The morphology of the arbor in M6 and of the sprig in M4 and M5 can be visualized in detail in ExM images (Figure 6A). The sprig is on a home column R7, where most of the synapses from yR7 to yDm8 are located (Figure 6B–C, Figure 6—figure supplement 1). In addition to the sprig, this yDm8 arbor has two thin dendritic processes that emerge from the base and are wrapped around neighboring non-home column R7 terminals (Figure 6C). The home column R7 extends into the base of the sprig, making multiple contacts with the base of the Dm8 arbor. DIP-γ mutant yDm8 neurons have thinner sprigs, in agreement with our quantitative analysis of mutant flipout clones from unexpanded samples (Figures 5I and 6D; Figure 6—figure supplement 2B).

Figure 6 with 6 supplements see all

Download asset Open asset

We previously showed that Dm8 cells are lost in DIP-γ^Mi>GFP/Df animals (Carrillo et al., 2015). To explore this in detail, we used CRISPR-generated null mutants for both DIP-γ and dpr11 (Xu et al., 2018) and determined yDm8 and pDm8 populations (Figure 7A–C). DIP-γ^Mi>GFP, which has no detectable protein expression, was used as one of the DIP-γ alleles to allow identification of yDm8 soma. Both DIP-γ and dpr11 mutants showed a ~ 50% decrease in yDm8 cell number using the pan-Dm8 driver and 60–65% loss when determined with Dac (Figure 7A,C). A double mutant had a similar phenotype to the two single mutants (Figure 7A), suggesting that the two genes function in the same pathway(s). The loss of yDm8 in DIP-γ mutants was partially rescued by expressing a DIP-γ transgene using the DIP-γ Gal4 driver (Figure 7A). To determine if yDm8 loss in dpr11 mutants was due to the absence of Dpr11 from the eye or from other Dpr11-expressing cells in the medulla, we performed eye-specific dpr11 transgenic RNAi, and found a ~ 40% reduction in yDm8 cell numbers (Figure 7C).

Figure 7 with 1 supplement see all

Download asset Open asset

We also ectopically expressed DIP-γ in PRs, and found that this phenocopied the DIP-γ and dpr11 loss-of-function (LOF) phenotypes (Figure 7C). This result suggests that the presence of both Dpr11 and its partner on the same cells (expression in cis) prevents Dpr11 in yR7 from interacting with DIP-γ in trans (on yDm8). Ectopic expression of DIP-α, a different DIP that does not bind to Dpr11, had no effect on yDm8 survival (Figure 7C), indicating that the Dpr11-DIP-γ interaction is required for yDm8 selection and survival. Similar results were observed for DIP-α and Dpr10 in the neuromuscular system, where DIP-α is normally expressed on the MNISN-1s (RP2) motoneuron and Dpr10 on its muscle targets. LOF mutants lacking either protein are missing specific muscle branches, and cis ectopic expression of DIP-α on muscles, or of Dpr10 on MNISN-1s, generates the same phenotypes (Ashley et al., 2019).

To assess whether the decrease in yDm8 population was due to apoptotic cell death, we expressed Drosophila inhibitor of apoptosis protein 1 (DIAP) in DIP-γ expressing cells (Figure 7—figure supplement 1). DIAP overexpression rescued the loss of yDm8 in DIP-γ mutants, indicating that DIP-γ is required for cell survival and that yDm8 undergo apoptosis when it is absent (Figure 7A). Furthermore, the number of yDm8 that survived when DIAP was expressed was ~45% greater than the number of yDm8 present in controls. It has been reported that extensive cell death occurs in the medullary cortex in wild-type during normal optic lobe development (Togane et al., 2012). Thus, in addition to the yDm8 that were rescued from cell death caused by the absence of DIP-γ, DIAP also rescued yDm8 that were lost due to normal developmental cell death (see Discussion). The fact that dpr11 and DIP-γ mutants displayed a similar extent of yDm8 loss implies that Dpr11 in yR7 signals to yDm8 via DIP-γ to ensure their survival. Absence of either molecule compromises the interaction and leads to death of yDm8 neurons.

We next assessed pDm8 populations in the above genotypes, including double mutants of dpr11 and DIP-γ, and found that they did not differ significantly from controls. Importantly, pDm8 cell numbers did not decrease when the yDm8 population was increased by DIAP rescue, indicating that the two populations of Dm8 neurons do not normally compete for survival signals (Figure 7B; see below).

We also visualized the M6 layer using the DIP-γ^Mi>GFP reporter and found that the morphologies of the yDm8 layer in the above genotypes were consistent with the results obtained by counting soma in the cortex. In the neuropil, larger gaps were observed in the mutants, and fewer and smaller gaps in wild-type and rescue genotypes (Figure 7D–H).

DIP-γ mutants also displayed a ~ 4.5-fold higher frequency of two-home column flipouts as compared to control (Figure 5G). The increase in two-home column yDm8 could be a response strategy in which yDm8 neurons innervate more yR7 PRs in order to compensate for the loss of survival signals mediated by DIP-γ.

Having found that interactions between yR7 and yDm8 mediated by Dpr11 and DIP-γ are required for yDm8 survival in adults (Figure 7), we next determined when the interaction that ensures survival occurs. To this end, we examined yDm8 loss at different stages of OL development in wild-type and DIP-γ mutant animals starting at 15 hr. (15 hr) after puparium formation (APF) (Figure 8A). yDm8 populations in wild-type and mutant animals were examined by labeling with Dac and DIP-γ^Mi>GFP reporter at 15 hr, 25 hr and 45 hr APF. At 15 hr APF, yDm8 cell numbers in DIP-γ mutants are similar to those in wild-type, suggesting that the Dpr11-DIP-γ interaction is needed for yDm8 survival only after that time. In wild-type, yDm8 numbers do not change between 15 hr APF and 25 hr APF. Between 25 hr APF and 45 hr APF, about 30% of yDm8 soma are lost, and yDm8 numbers then remain stable until adulthood (Figure 8A). However, in DIP-γ mutants, more than 1/3 of the yDm8 soma are lost between 15 hr APF and 25 hr APF, and by 45 hr APF only 25% of the original number of yDm8 soma present at 15 hr APF remain (Figure 8A). As in wild-type, these numbers in DIP-γ mutants then remain stable until adulthood. Inhibition of cell death in the DIP-γ mutant with DIAP restored the population in the adult back to the original wild-type pool size present at 15 hr APF (Figure 8A). Thus, DIP-γ is required for suppression of yDm8 cell death early in pupal development. Extensive migration of yDm8 soma was observed in wild-type, and this migration was unaffected in the DIP-γ mutant (Figure 8—figure supplement 2A–D).

Figure 8 with 2 supplements see all

Download asset Open asset

To detect cells that are undergoing apoptosis in DIP-γ mutants, we used Apoliner, a fluorescent reporter of caspase activity, to mark yDm8 at 25 hr APF in wild-type and mutant genotypes (Bardet et al., 2008) (Figure 8B–E). The Apoliner reporter consists of a membrane tethered RFP joined to GFP with a nuclear localization signal and a caspase site between the RFP and GFP moieties. In cells undergoing apoptosis, activated caspases cleave Apoliner, releasing GFP to localize to the nucleus (Figure 8B). The Apoliner transgene was driven by a DIP-γ split Gal4 driver in wild-type and DIP-γ mutant and yDm8 cell bodies were identified by labeling with Dac. At 24–25 hr APF, there were twice as many dying yDm8 cells in DIP-γ mutants as compared to control, further confirming the role of Dpr11-DIP-γ interaction in yDm8 cell survival (Figure 8C–E).

We next determined when Dpr11-expressing yR7 and DIP-γ expressing yDm8 meet in the medulla neuropil. Since yDm8 cell death caused by the loss of DIP-γ begins to occur between 15 hr and 25 hr APF (Figure 8A–E) and because R7 terminals do not mature till after the mid-pupal stage (Ting et al., 2005), we investigated if the Dpr11-DIP-γ interaction required for yDm8 survival occurred at the growth cone stage of R7 development. Indeed, we found that dpr11 was expressed in select R7 growth cones at 15 hr APF, suggesting that the yellow fate of R7 was determined by that time (Figure 8F). We next examined DIP-γ expression in early pupa and found DIP-γ labeling adjacent to select R7 at 20 hr APF, indicating that yDm8 arbors are apposed to yR7 terminals (Figure 8G). The gaps in DIP-γ labeling are presumed to be pDm8 (Figure 8—figure supplement 1). Taken together, our results are consistent with a model in which selection of yDm8 by yR7 occurs by 15 hr - 20 hr APF via Dpr11-DIP-γ binding and thereby ensures survival of yDm8.

Since Dpr11 is expressed in yR7 PRs, we next examined how the two subtypes of Dm8 neurons respond when R7 fates are altered. The transcription factor Spineless (Ss) controls yellow and pale subtype choice in the retina. Ectopic expression of ss in all PRs (ss^GOF) induces yellow fate in all R7 cells and prevents formation of pR7 (Wernet et al., 2006). We first determined Dpr11 and DIP-γ expression in ss^GOF. In contrast to wild-type, where only yR7 is labeled by dpr11^Mi>GFP, all R7 terminals showed GFP expression when ss was mis-expressed (Figure 9A–B). Thus, Ss expression activates Dpr11 along with Rh4 and other yR7-specific genes. The normal gaps in DIP-γ^Mi>GFP labeling in the M6 layer were missing in ss^GOF (compare 1A and C in Figure 9—figure supplement 1). When DIP-γ was removed in the ss^GOF background, gaps were again observed in the yDm8 layer, indicating that the M6 signal in ss^GOF is indeed due to DIP-γ-expressing yDm8 (Figure 9—figure supplement 1A–C). The change in R7 fate is thus transmitted to the downstream circuit in the medulla via Dpr11-DIP-γ interactions, resulting in uniform yDm8 labeling in the M6 layer.

Figure 9 with 2 supplements see all

Download asset Open asset

We determined yDm8 and pDm8 cell number in the above genotypes (Figure 9C–D). In ss^GOF, we observed a 36% increase in yDm8, while pDm8 cell number was decreased by 84%. The increase in yDm8 in ss^GOF was dependent on DIP-γ, as absence of DIP-γ in the ss^GOF background resulted in 53% loss of yDm8 neurons. Interestingly, pDm8 increased significantly, doubling their cell numbers as compared to ss^GOF alone (Figure 9D). Thus, yDm8 and pDm8 populations compete with each other when R7 fate is altered.

To assess Dm8 subtypes in the reverse situation where there are only pR7 in the eye, we ectopically expressed the Defective proventriculus (Dve) transcription factor, a repressor downstream of ss, in all PRs (Johnston et al., 2011; Nakagawa et al., 2011; Yan et al., 2017). Expression of Dve in all PRs (dve^GOF) converts all ommatidia to pale (Yan et al., 2017). We used this strategy because ss LOF mutations cause lethality and we observed no phenotype with eye-specific ss RNAi. Since the external eye morphologies of dve^GOF animals were abnormal, we initially examined the neuropil layers and PR targeting in the medulla (Figure 9—figure supplement 1E,K). Except for the DIP-γ^Mi>GFP labeling in M6, other layers of the neuropil showed similar patterns to wild-type. The M6 layer had large gaps, similar to those observed in DIP-γ mutants, suggesting that there were fewer yDm8 neurons (Figure 9—figure supplement 1E). Indeed, yDm8 cell number decreased by ~60% as compared to the control (Figure 9G).

Sevenless is a receptor tyrosine kinase that is required for specification of R7 in the developing eye. In sev mutants, R7 PRs are absent, because the R7 precursor cell becomes a non-neuronal cone cell. We found that both the yDm8 and pDm8 populations were reduced in sev mutants by >70% as compared to control cell numbers, showing that both classes of Dm8 require interaction with R7 PRs for survival. yDm8 cell numbers were further reduced when DIP-γ was removed in the sev mutant background (Figure 9C). This dependence on DIP-γ likely occurs because there are a few R7 remaining in the sev mutant (Figure 9—figure supplement 1F), although the allele we used is described as an amorph. The ~50 yDm8 that remain in a sev DIP-γ double mutant may not require interactions with other cells for survival, or may obtain support through a different pathway that does not involve DIP-γ. pDm8 numbers are slightly increased by removal of DIP-γ in the sev mutant background. This is probably because they are competing with yDm8 for the few remaining R7 PRs, so that the loss of additional yDm8 through removal of DIP-γ frees up those slots to be occupied by pDm8.

To complete our analysis of the consequences of changing R7 fates, we examined ectopic expression of Dpr11 in the eye (dpr11^GOF). We reasoned that since dpr11 was expressed in all R7 in ss^GOF (Figure 9B), Dpr11 overexpression in the eye should mimic ss^GOF. Indeed, the M6 layer showed continuous yDm8 labeling with no pDm8 gaps, similar to ss^GOF (Figure 9—figure supplement 1D). This was confirmed by determining yDm8 and pDm8 cell numbers in dpr11^GOF. The yDm8 population increased by 53% over control when Dpr11 was overexpressed in all PRs, accompanied by a 67% loss of pDm8 (Figure 9C–D). Dpr11 overexpression in dpr11^GOF was confirmed by labeling with Dpr11 antibody. Dpr11 overexpression does not convert pR7 to the yR7 fate, because R7 PRs that ectopically express Dpr11 do not all express Rh4 (Figure 9—figure supplement 1G–I).

We next examined cross-sectional (top-down) views of the mid-pupal medulla in dpr11^GOF labeled with DIP-γ and Chp antibodies to assess the composition of Dm8 columns when Dpr11 is expressed in all R7 PRs. Wild-type controls showed 62y:38p columns in the medulla, similar to the 65y:35p ratio of yellow and pale ommatidia in the eye. By contrast, in dpr11^GOF only yellow columns were observed, indicating that all pDm8 home columns had been replaced by yDm8 (Figure 9E–F). Thus, ~35% of yDm8 neurons had now selected PRs that express Dpr11 but were otherwise of the p subtype as their home column R7 (Figure 9H, Figure 9—figure supplement 1I). Similar results were obtained in ss^GOF, in which pR7 PRs are actually converted to the y subtype (Figure 9—figure supplement 1J). Interestingly, we did not observe any mistargeting of yDm8 to the M3 layer in dpr11^GOF, although Dpr11 was overexpressed in all PRs including R8 (Figure 9—figure supplement 1L).

Since ectopic expression of Dpr11 in all PRs changed the identities of home column Dm8, converting them all to yDm8 (Figure 9E–F,H), we investigated whether there were defects in home column selection in dpr11 LOF mutants. We examined mutants and heterozygous controls at 44 hr APF because Dm8 have not yet contacted neighboring columns at this stage, making it possible to unambiguously assign the home column (Ting et al., 2014). y and p R7 were identified using dpr11^Mi>GFP and Chp antibody. Using the DIP-γ antibody to label yDm8, we determined how many yR7 containing columns were paired with yDm8. In controls, every yR7 column had DIP-γ labeling adjacent to it. In dpr11 mutants, there was a significantly higher percentage of unpaired yR7 as compared to controls (Figure 9—figure supplement 2D). However, this result can be explained by the fact that yDm8 die in dpr11 mutants, so that the number of surviving yDm8 neurons is insufficient to innervate all yR7. Thus, the absence of DIP-γ labeling adjacent to yR7 does not indicate that Dpr11 is required for home column selection. We found no mistargeting errors where pR7 containing columns labeled only with Chp were apposed to yDm8 in dpr11 mutants (Figure 9—figure supplement 2A–B).

We repeated the above experimental paradigm in a genotype where yDm8 cell death was inhibited in a dpr11 mutant and found that this eliminated unpaired yR7. We observed very rare mistargeting of a yDm8 to a pR7 in this genotype (2/119 pR7 (or 300 total R7 PRs) examined) (Figure 9—figure supplement 2C). We hypothesize that pR7 may not be available as yDm8 targets in dpr11 mutants because they would have been selected by pDm8 neurons. The fact that pR7 are occupied by pDm8 makes it difficult to assess whether Dpr11-DIP-γ interactions are required for selection of yR7 as targets by yDm8.

Yellow and pale ommatidia are generated by a stochastic process and are distributed in a ~ 65y:35p ratio in the retina (Wernet et al., 2006). Since yDm8 and pDm8 are separate populations (Figure 2) and R7 y vs. p fates are randomly determined, what mechanisms ensure that each yDm8 has a yR7 partner and each pDm8 has a pR7 partner? The strategy appears similar to those used in the development of many mammalian nervous system circuits, in that yDm8 neurons are generated in excess of their final cell numbers in adults, and those that are not selected for survival by an appropriate home column R7 partner die through apoptosis.

Our conclusions are summarized in a model (Figure 10). There are two subtypes of Dm8: yDm8 that express DIP-γ and pDm8 that do not. yDm8 are born in excess of the final cell numbers that are present in the adult (Figure 8A). There are likely to be extra pDm8 born as well, but due to the lack of a specific pDm8 marker we cannot determine their cell numbers in early pupae. Dpr11, the binding partner of DIP-γ, is expressed exclusively in yR7 in the retina (Figure 1E–F), allowing yDm8 neurons to select yR7 PRs as their appropriate partners. The extra yDm8 that do not find a yR7 partner undergo cell death due to loss of survival signals mediated by DIP-γ-Dpr11 interactions. This selection mechanism ensures that the ratio of yDm8 to pDm8 in the medulla matches the ratio of yR7 to pR7 in the eye (Figure 1I).

Figure 10

Download asset Open asset

Dpr11 and DIP-γ are required for survival, because 50–60% of yDm8 die when either molecule is absent (Figures 7, 8A and 10). If all R7 are converted to yR7 (in ss^GOF), or if they all express Dpr11, many more yDm8 neurons survive than in wild-type, showing that the numbers of Dpr11-expressing R7 are limiting for yDm8 survival in wild-type animals (Figures 9C and 10). This seems surprising, because a wild-type retina should contain 460–500 yR7 (Posnien et al., 2012), which is greater than the number of yDm8 we count in early pupae (Figure 8A). However, the actual yDm8 cell numbers may be larger due to incomplete penetrance of the markers, as discussed in Results. yDm8 and pDm8 populations do not compete with each other in wild-type because they have different R7 subtype partners. When the numbers of yDm8 neurons are increased by suppressing apoptotic cell death with DIAP in yDm8, pDm8 cell numbers do not decrease (Figure 7B). However, yDm8 and pDm8 neurons can be forced to compete by changing R7 subtype fate. When all R7 PRs are converted to yR7, or when they all express Dpr11, almost all pDm8 neurons die (Figures 9D and 10). Conversely, most yDm8 neurons die when all R7 PRs are converted to pR7 (Figure 9G). pDm8 neurons are preferentially selected by pR7 PRs, using unknown molecular mechanisms, and pR7 PRs are needed to ensure pDm8 survival. This is demonstrated by the fact that pDm8 are unable to effectively fill vacant yR7 slots. For example, in ss^GOF, DIP-γ animals, there are at least 180 yR7 slots that are made vacant by loss of yDm8, but the number of pDm8 neurons only increases by ~30 (~2 fold) in this genotype (Figure 9C–D).

Dpr11 is already expressed in a subset of R7 PRs at 15 hr APF, indicating that yR7 fate is determined by that time (Figure 8F). DIP-γ is expressed by 18 hr - 20 hr APF in select Dm8 neurons apposed to Dpr11-expressing R7 PRs (Figure 8G; Figure 8—figure supplement 1A–B). At this time, R7 growth cones are positioned in the incipient layer (Kulkarni et al., 2016; Ting et al., 2005). This is consistent with earlier findings that single-cell Dm8 clones contact home column R7 growth cones by 17 hr APF (Ting et al., 2014). Taken together, these results indicate that yR7 and yDm8 have met and selected each other in the incipient layer by 15 hr - 20 hr APF. Survival signaling mediated by Dpr11-DIP-γ interactions is probably initiated around that time, because cell death of yDm8 in DIP-γ mutants begins to occur between 15 hr and 25 hr APF (Figure 8A).

There is widespread cell death in the medullary cortex as part of normal optic lobe development. This occurs in two phases, with the first phase spanning 0 hr - 48 hr APF and peaking at 24 hr APF (Togane et al., 2012). In wild-type, ~30% of the yDm8 neurons present at 15 hr APF die by 45 hr APF, at which time yDm8 cell numbers are the same as those in adult. In DIP-γ mutants, ~75% are lost by 45 hr APF. If apoptotic cell death is suppressed by DIAP, then the yDm8 population in adult remains the same as at 15 hr APF (Figure 8A). We suggest that the yDm8 neurons that die in wild-type are those that are in excess of the number that can be selected for survival by Dpr11-expressing yR7 PRs. Since most yDm8 neurons require DIP-γ for survival, we infer that the excess yDm8 neurons in wild-type die because they lack survival signals that are transmitted through Dpr11-DIP-γ interactions. These selection events are independent of synaptogenesis between R7 and Dm8, which occurs only after 60 hr APF.

The DIP-α-expressing Dm4 neurons are also generated in excess, and their survival is regulated by interactions between DIP-α and its partners Dprs 6 and 10, which are expressed on presynaptic L3 neurons. In DIP-α or dpr6 dpr10 mutants, about 50% of Dm4 and 20% of Dm12 neurons are lost. If cell death is blocked in wild-type, the cell numbers of Dm4 neurons in adults are increased by about 25% (Xu et al., 2018). Thus, in these cases, Dpr-DIP interactions play a similar role in suppressing cell death of postsynaptic neurons in the distal medulla. However, not all DIP-γ expressing neurons in the optic lobe require it for survival. A subset of lobula plate tangential cells (LPTCs) also express DIP-γ, and these do not die in DIP-γ mutants (unpublished data).

Our gain-of-function studies show that when Dpr11 is expressed in all R7, all pDm8 home columns are converted to yDm8 (Figure 9E–F,H), demonstrating that expression of Dpr11 in R7 PRs that are otherwise of the pale subtype is sufficient for selection of yDm8. However, we were unable to determine whether Dpr11 is also necessary for yDm8-yR7 pairing by analysis of LOF mutants. We did not observe mistargeting of surviving yDm8 in dpr11 mutants to pR7 columns, most likely because pR7 slots are occupied by pDm8 neurons (Figure 9—figure supplement 2A–D). Another factor that could contribute to accurate targeting in dpr11 mutants is that there may be a second mechanism for support (and possibly selection) of yDm8 by R7 that is independent of Dpr11-DIP-γ interactions. This is suggested by the fact that ~70% of yDm8 are lost in sev mutants, but only ~50% in DIP-γ and dpr11 mutants (Figures 9C and 7A).

A typical wild-type yDm8 arbor has a thick distal dendritic projection (sprig) that extends along the home column yR7 terminal, usually reaching M4. In DIP-γ and dpr11 mutants, the distribution of sprig diameters for yDm8 neurons is shifted toward smaller values (Figure 5I), suggesting that mutant sprigs may have a reduced capacity to form synapses with yR7 in M4 and M5. Synapses that would have been in M4 and M5 might then be found in the base of the arbor in M6. EM reconstruction data for columns D and E supports this idea. R7 output synapses (T-bars) are mainly in M4 and M5 (on the sprig) in column E, which has a Dm8 with a thick sprig (Figures 2A and 4A, inset in 4B). In column D, whose Dm8 has a very thin sprig, R7 T-bars are located in M6 (in the main arbor) (Figure 4A, inset in 4C).

The effects of Dpr11 and DIP-γ on yDm8 sprig morphology may involve a different pathway from that involved in cell death. Although cell death would have occurred in the DIP-γ and dpr11 mutant animals, these flipouts were analyzed in adults, in which the population has stabilized and the remaining yDm8 neurons are not in the process of dying. Sprig morphology defects might be a consequence of a loss of Dpr11-DIP-γ-mediated adhesion between yR7 and yDm8, since dpr11 and DIP-γ mutants are similarly affected.

We had earlier reported an ‘overshoot’ phenotype in dpr11^Mi>GFP/Df mutants in which some R7 terminals had processes extending beyond M6 into proximal medulla layers (Carrillo et al., 2015). Overshoots were labeled by an overexpressed truncated form (Brp-short) of the active zone protein Bruchpilot (Brp) (Berger-Müller et al., 2013). However, this phenotype was not observed in the dpr11^CRISPR null mutant when endogenous Brp puncta were labeled with the STaR method (Xu et al., 2018). To determine whether this discrepancy was due to the dpr11^Mi>GFP/Df genotype or to the Brp-short marker, we analyzed dpr11^CRISPR homozygotes using Brp-short and observed a significant increase in yR7 overshoots as compared with the heterozygous control (Figure 9—figure supplement 2E). However, the interpretation of this Brp-short phenotype is unclear, as it does not correspond to a shift in the distribution of endogenous Brp puncta into layers proximal to M6.

Tm5a, Tm5b, and Tm5c are output neurons for R7 and R8 circuits. Their dendrites receive direct synaptic input from R7, R8, and Dm8, and their axons relay signals to the lobula (Gao et al., 2008; Karuppudurai et al., 2014; Takemura et al., 2013; Takemura et al., 2015). It has been hypothesized that Tm5a/b/c cells are analogous to vertebrate retinal ganglion cells, and that true color vision (intensity-independent hue discrimination) involves interpretation of Tm inputs by circuits in the lobula. Chi-Hon Lee and his colleagues developed a learning paradigm in which flies discriminate between equiluminant blue and green light, which are differentially sensed by y (Rh6) and p (Rh5) R8. Four classes of Tm neurons (Tm5 a/b/c and Tm20) make redundant contributions to this response (Melnattur et al., 2014).

The absorption spectra of Rh4 (yR7) and Rh3 (pR7) extensively overlap (Figure 11C), and it is unlikely that a learned color vision paradigm could be developed that would distinguish between Rh4 and Rh3 inputs. Nevertheless, the existence of discrete y and p R7 channels suggests that the fly utilizes them to make discriminations among UV wavelength inputs. To make such discriminations would require that the y and p channels have different synaptic circuits with distinguishable outputs. Our analysis of the patterns of synaptic connections defined by the EM reconstruction of the medulla (Takemura et al., 2013; Takemura et al., 2015) indicates that yR7 and yDm8 both preferentially synapse onto Tm5a, while pR7 and pDm8 synapse onto Tm5b (Figure 3). Tm5a expresses DIP-γ, while Tm5b does not (Cosmanescu et al., 2018), so Dpr11-DIP-γ interactions might be involved in specifying both yR7-yDm8 and yR7-Tm5a connections. Four of the seven Dm8 neurons, and three of the eight R7 PRs, also make a few synapses onto Tm5c neurons, which are required for innate UV preference (Karuppudurai et al., 2014) (Figure 3—source data 1). There is no specificity for y vs. p in the connections to Tm5c.

Figure 11

Download asset Open asset

We visualized yR7-yDm8-Tm5a (yellow) and pR7-pDm8-Tm5b (pale) circuits using 3D renderings of cells from the EM reconstruction (Takemura et al., 2015) (Figure 4A–C and associated videos). The yDm8-E sprig and Tm5a-E dendritic branch are both wrapped tightly around the yR7-E terminal, and most yR7 T-bars are apposed to the sprig and branch within layer M5. Dm8 arbors contain both pre- and postsynaptic elements, and therefore both receive inputs and emit outputs. yDm8-E T-bars are distributed between the sprig and the main arbor in M6 (Figure 4B). Remarkably, our 3D renderings of wild-type yDm8 and yR7 neurons from ExM (Figure 6 and associated videos) are almost superimposable on yDm8-E and yR7-E from the EM reconstruction (Figure 6—figure supplement 1). This allows us to interpret yDm8 ExM phenotypes in DIP-γ mutants (Figure 6, Figure 6—figure supplement 2) by reference to the EM reconstruction.

R7 is histaminergic, and inhibits both Dm8 and Tm5a/b (Figure 11A–B), usually through polyadic synapses where a single R7 T-bar is apposed to both a Dm8 and a Tm5a/b postysnapse (Figure 4D–E). Dm8 is glutamatergic, while Tm5a and Tm5b are cholinergic (Davis et al., 2018; Gao et al., 2008; Karuppudurai et al., 2014). This suggests that yR7 input might cause inhibition of Tm5a in two ways: by direct inhibition and by inhibiting the glutamatergic input (probably excitatory) of yDm8 onto Tm5a. Conversely, pR7 input could preferentially inhibit Tm5b by both direct and indirect mechanisms. There might be timing differences between direct and indirect inhibition.

UV inputs will always stimulate both y and pR7 channels, because the Rh4 and Rh3 absorption maxima differ by only 20 nm. However, longer-wave inputs that activate Rh4 on yR7 more than Rh3 on pR7 might cause more inhibition of Tm5a than of Tm5b, while the reverse would be true of shorter-wave inputs that preferentially activate Rh3 on pR7 (Figure 11). Combining direct (R7→Tm5a/b) and indirect (R7→Dm8→Tm5a/b) inhibition of Tm5a and Tm5b outputs might amplify these effects, depending on the relative timing of R7 and Dm8 inputs. This model suggests that neurons in the lobula or elsewhere that can read the ratio of Tm5a to Tm5b output mediate UV wavelength discrimination. Tm5a/b/c and Tm20 as a group synapse onto many lobula neuron types, but the specific partners of Tm5a and Tm5b are mostly unknown. Interestingly, however, all Tm5a but only half of Tm5b neurons synapse onto LT11 lobula projection neurons (Lin et al., 2016). Flies with silenced LT11 neurons have reduced phototaxis toward blue light (Otsuna et al., 2014).

In conclusion, our results show that Dpr11 and DIP-γ expression patterns define a yR7-yDm8-Tm5a color vision circuit that should preferentially respond to longer-wavelength UV input. Suppression of yDm8 cell death triggered by engagement of Dpr11 on yR7 by DIP-γ on yDm8 helps to build this circuit by ensuring that each yDm8 has a home column yR7.

Genotypes of all the flies analyzed in this study are indicated in Tables 2–3 and lines obtained from other sources are listed in Table 1. Heterozygote controls were used in all experiments for determining cell numbers. DIP-γ^Mi>GFP reporter (indicated in all graphs as DIP-γ ^-/+) is an insertion in the 5’ UTR intron and has reduced or no protein expression. Thus, this line serves as a mutant as well as a reporter of DIP-γ transcript.

For eye-specific transgenic RNAi, we screened three dpr11 RNAi lines by crossing to a line which had one copy of dpr11 removed to increase the effectiveness of the RNAi line: lGMR-Gal4; dpr11^null, DIP-γ^MI03222-GFP; the RNAi line GD2343 (VDRC) had the strongest phenotype and was used in the paper.

Generation of UAS-transgenic flies cDNA encoding Dpr11 and DIP-γ were cloned from pOT2 GH22307 for Dpr11 and pOT2 GH08175 for DIP-γ into pUAST attB vector using standard molecular biology techniques. 5’ UTRs for both genes contained many upstream ATG codons; we made deletions of the 5’ UTRs so that the ATGs of the proteins were the first ATGs in the mRNAs, in order to increase expression of the transgenes (sequences available on request). Transgenes were injected into embryos (Rainbow Transgenics). UAS-Dpr11 and UAS-DIP-γ both used the attP40 (2L) landing site.

The primary antibodies used were as follows: anti-DIP-γ (guinea pig, 1:200) and anti-Dpr11 (rabbit, 1:150) were gifts from C. Desplan. Anti-Pros MR1A (mouse 1:4), anti-Elav 7E8A10 (rat, 1:10), anti-Dac 2–3 (mouse 1:50), anti-chaoptin 24B10 (mouse 1:20) were obtained from Developmental Studies Hybridoma Bank (University of Iowa, IA). Commercial antibodies were used as follows: Rabbit anti-RFP (Rockland, 1:500), rabbit anti-GFP (Thermo- Fisher Scientific, 1:500), chicken anti-GFP (Aves labs, 1:500), mouse anti-myc 9E10 (Abcam, 1:500) and chicken anti-beta-galactosidase (Abcam, 1:1000). Secondary antibodies were obtained from Thermo-Fisher Scientific and used at 1:500.

Eclosed flies (less than 3 days old) were dissected in phosphate-buffered saline (PBS) and fixed in 4% paraformaldehyde in PBS with 0.2% Triton-X-100 (PBT) for 20 min at room temperature. Brains were washed overnight in PBT, followed by a two-day incubation at 4°C with primary antibody that was diluted in blocking buffer (5% normal goat serum in PBT). Samples were then incubated with secondary antibody (similarly diluted in blocking buffer) for two-days, followed by washing with PBT and PBS and stored in Vectashield.

For single cell flipouts, brains were given heat-shock at 50 hr APF for 10–20 min at 37°C and 1 day eclosed flies were dissected for staining as above.

For cell counts and column analyses, optic lobes were separated from the central brain and mounted top-down. Confocal images were acquired on Zeiss LSM800 or LSM700 microscopes with a 40xobjective. For cell counts and single-cell flipouts, optic lobes were imaged with 1.5 µm and 0.8 µm z-sections, respectively. Single slices or maximum intensity projections were exported with Zen software (Zeiss) for image processing.

Images were processed with Adobe Photoshop.

Cell numbers were determined blind to genotype using Zen software to do manual counts. 1-3d old adults were used for counts in adults. For all data (cell soma, columns and Brp overshoots), only one optic lobe per animal was used for counts and statistics. p-value was determined using Student’s unpaired t-test from Graphpad Prism. All data reported in graphs are mean + /- standard deviation. yDm8 sprig parameters were measured using Image J software. To obtain measurement of the widest point of a sprig, a slice at which the image was most in focus was selected. The height was measured from the most distal point of the sprig to its most proximal point before the processes of the sprig connected back to the dendritic base. The Brp overshoots were analyzed blind using Image J software, by an individual unrelated to that experiment.

For quantitation of neighboring contacts in all genotypes, well isolated clones were chosen for this analysis. The presence of a single axonal process leading to a soma confirmed that each clone was indeed a single cell. Images were then viewed in Imaris. The identity of the home column was excluded from the yellow vs. pale R7 contacts.

Single-cell flipouts were processed for expansion microscopy after immunohistochemistry (and confirming the presence of flipouts) according to a protocol from Tim Mosca, based on Mosca et al. (2017). Briefly, labeled brains were washed in 50% PBT-PBS mixture before proceeding with steps for anchoring proteins to the sodium acrylate matrix. Gelling chambers were made on coated slides with No.1 coverslips as bridge and brains were embedded in sodium acrylate gel. After incubation for 2 hr at 37°C, the samples were excised from the solidified gel and processed for digestion with proteinase K in six-well plates. Gel fragments were expanded by replacing the digestion buffer with several washes of water. Finally, in preparation for confocal imaging, the gel fragments containing the expanded transparent brains were placed on a 24 mm x 50 mm No. 1.5 coverslip that was coated with poly-l-lysine. Preps were always covered in water to make sure they did not dry out during imaging. Expanded brains were imaged on Zeiss LSM800 with 40x Objective 1.1 NA.

ExM images were processed for 3D reconstruction using Imaris software (Bitplane). Using surface rendering and fluorescence thresholding, we were able to obtain an accurate 3D structure of a yDm8 in wild-type and DIP-γ mutant. Each channel was recreated by adding an additional surface. The threshold for new surface detection was set as needed per image minimizing the addition of surface fragments not present in the original image. Each new surface was segmented in order to isolate the appropriate region of interest. Video recordings were captured at 10 frames per second for a total of 300 frames.

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

We thank Larry Zipursky, Aljoscha Nern, Takashi Suzuki, and Michael Reiser for fly lines and discussions, Maximilien Courgeon and Claude Desplan for sharing unpublished antibodies and fly lines and for discussions, and Tim Mosca for his ExM protocol. We thank Violana Nesterova for figure preparation, Shuwa Xu and Larry Zipursky for comments on the manuscript, Shuwa Xu for discussions during the course of the work, and Namrata Bali for help with ExM. The dpr11 RNAi stock was obtained from the Vienna Drosophila Resource Center (VDRC, www.vdrc.at). Stocks obtained from the Bloomington Drosophila Stock Center (NIHP40OD018537) were used in this study. Imaging was performed in the Biological Imaging Facility, with the support of the Caltech Beckman Institute and the Arnold and Mabel Beckman Foundation. Anti-Pros MR1A (mouse 1:4), anti-Elav 7E8A10 (rat, 1:10), anti-Dac 2–3 (mouse 1:50), anti-chaoptin 24B10 (mouse 1:20) were obtained from the Developmental Studies Hybridoma Bank (University of Iowa, IA). This work was supported by NIH grants RO1 EY028116 and R37 NS28182 to KZ, and by the Howard Hughes Medical Institute (S-YT).

© 2019, Menon et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.