Two distinct mechanisms of Plexin A function in Drosophila optic lobe lamination and morphogenesis Free

During nervous system development, interactions between cell-surface molecules determine the formation of synaptic connections, which are often concentrated in synaptically dense regions called neuropils. Spatial segregation of the synaptic connections between different neuronal types, e.g. in layers or nuclei, simplifies the process of growth cone navigation to the correct target cells (Sanes and Zipursky, 2010). Many regions of vertebrate and invertebrate nervous systems, including the Drosophila optic lobe (Millard and Pecot, 2018; Sanes and Zipursky, 2010), the vertebrate retina (Baier, 2013), and the mammalian cortex and cerebellum (Guy and Staiger, 2017), are divided into synaptic layers in which neurites of distinct cell types form localized connections with specific synaptic partners.

Several mechanisms that establish this laminated organization have been identified in the vertebrate retina. Formation of synaptic laminae in the inner plexiform layer of the chick retina depends on differential localization of Sidekicks, Contactins and Dscams, which are members of the immunoglobulin (Ig) superfamily, in non-overlapping groups of bipolar, amacrine and ganglion cells (Yamagata and Sanes, 2008, 2012). Homophilic interactions between these molecules direct neurites that express the same Ig protein to the same layer (Yamagata et al., 2002). Another mechanism that organizes sublaminae in the mouse inner plexiform layer is complementary localization of receptor-ligand pairs; localization of Semaphorin 6A in ON laminae, of Plexin A4 in OFF laminae and of Plexin A2 in both establishes the stratification of the motion detection circuit via repulsive signaling (Matsuoka et al., 2011; Sun et al., 2013). Studying a simpler model system with fewer gene duplications may reveal additional insights into layer formation.

The Drosophila medulla, the largest of the four optic lobe neuropils, resembles the inner plexiform layer of the vertebrate retina (Sanes and Zipursky, 2010). Its ten layers, which represent highly stereotyped and segregated neurite projections from numerous cell types, emerge over the course of pupal development (Ngo et al., 2017). Although Drosophila Sidekick (Sdk) localizes to distinct layers of the medulla neuropil, loss of sdk does not disrupt medulla layers (Astigarraga et al., 2018). As in the mouse retina, lamination in the fly medulla requires complementary localization of plexins and semaphorins. Lamina neurons receive input from motion-sensitive photoreceptors and form output synapses in specific layers of the outer medulla neuropil. Repulsive interactions between Semaphorin 1a (Sema1a) expressed in lamina neurons and Plexin A (PlexA) expressed by tangential neurons in the M7 layer of the medulla prevent lamina neurons from projecting into the inner medulla (Pecot et al., 2013). Thus, subdivision of neuropils into functional synaptic laminae is a common organizational strategy that involves conserved molecular mechanisms.

PlexA is one of two Drosophila members of the plexin family of transmembrane signaling proteins. Plexins interact with their canonical partners, semaphorins, through the extracellular Sema domains of both proteins (Andermatt et al., 2014; Janssen et al., 2010; Liu et al., 2010; Nogi et al., 2010; Winberg et al., 1998). Plexin A proteins act as receptors for class I transmembrane semaphorins during axon guidance in vertebrates and invertebrates (Hung et al., 2010; Negishi et al., 2005; Tamagnone et al., 1999), and signal through an intracellular RasGAP domain, as well as interacting with other downstream effectors (Ayoob et al., 2004; Hung et al., 2010; Pascoe et al., 2015). Overexpression studies suggest that Plexin A proteins are also ligands for a Semaphorin 1 reverse signaling pathway that remodels the actin cytoskeleton (Andermatt et al., 2014; Jeong et al., 2012, 2017; Sun et al., 2015; Toyofuku et al., 2004; Yu et al., 2010).

Here, we show that PlexA establishes synaptic layering in the Drosophila medulla during pupal development. Loss of PlexA disrupts neuronal projections throughout the medulla, resulting in global disorganization of medulla layers. We use CRISPR modification of the endogenous PlexA gene to show that PlexA mediates layering independently of its cytoplasmic domain, likely serving as a ligand for Sema1a in specific medulla neuron types. In contrast, PlexA requires its cytoplasmic domain to establish the overall shape of the medulla neuropil during pupal development. We propose that PlexA has dual functions during medulla development, as a receptor to control neuropil morphogenesis, and as a ligand for Sema1a to establish neurite segregation that gives rise to a laminated medulla architecture.

Targeting of color-sensitive R7 and R8 photoreceptor axons to distinct layers of the Drosophila medulla is a powerful model with which to investigate the organization of synaptic layers. Several studies have identified factors that act in the photoreceptors themselves to enable R8 to terminate in the M3 layer of the medulla and R7 to project beyond it to the M6 layer (Astigarraga et al., 2010; Plazaola-Sasieta et al., 2017). However, few of the molecular cues that photoreceptor axons recognize within these medulla layers are known (Timofeev et al., 2012). To identify such cues, we carried out an RNAi screen in which we knocked down genes predicted to encode cell-surface and secreted molecules (Kurusu et al., 2008) with a combination of three GAL4 drivers [apterous (ap)-GAL4, Distal-less (Dll)-GAL4 and eyeless (ey^OK107)-GAL4] that together are expressed in a large proportion of medulla neurons (Li et al., 2013; Morante and Desplan, 2008) (Fig. S1A). We found that knocking down PlexA caused R7 axons to terminate prematurely in the same layer as R8 axons in the anterior medulla (Fig. 1A,B). A second RNAi line targeting PlexA had the same phenotype (Fig. 1C; Fig. S1B,C), ruling out an off-target effect.

PlexA is a transmembrane protein that contains an N-terminal Sema domain, three plexin, semaphorin and integrin (PSI) domains, and three immunoglobulin-like, plexin and transcription factor (IPT) domains in the extracellular region, and an intracellular bipartite RasGAP domain (Fig. 1D). To confirm that PlexA is required for normal photoreceptor axon projections, we used CRISPR mutagenesis to delete the region that encodes the transmembrane domain and much of the extracellular domain (Fig. 1C,D), generating a protein null allele, PlexA²² (Fig. 1E). We also examined PlexA^MB09499 (Bellen et al., 2011), a Minos transposable element insertion into a coding exon (Fig. 1C). Homozygotes for either mutation die before eclosion. In PlexA homozygous mutant pupae, R7 and R8 projections were disordered and did not target distinct layers in the medulla (Fig. 1F-H). Additionally, PlexA mutant photoreceptor axons were misshapen and formed less distinct terminals than controls (Fig. 1F-H).

PlexA is expressed in the eye as well as the brain, and one of the GAL4 lines we used for our screen, ey^OK107-GAL4, also drives expression in the eye. To test whether these defects were caused by loss of PlexA in photoreceptors, we generated homozygous PlexA mutant clones in the eye using a duplication of the PlexA genomic region inserted on the third chromosome. R7 photoreceptor axons terminated correctly in the M6 layer, even in very large PlexA mutant clones generated in a Minute background (Fig. S1D). This indicates that PlexA is not required autonomously in photoreceptor cells, but acts in medulla neurons to drive normal photoreceptor axon targeting.

During mid-pupal development, R7 cells form synapses with Dm8 interneurons and other post-synaptic partners in the M6 layer (Gao et al., 2008; Kind et al., 2021; Menon et al., 2019). Mistargeting of photoreceptor axons might prevent Dm8 neurites from contacting photoreceptor axon terminals, precluding these synapses from forming. We labelled Dm8 cells with UAS-myrTomato driven by ort^C2b-GAL4 (Gao et al., 2008) in PlexA mutant flies to determine their position relative to R7 axons. In control pupae, Dm8 dendrites arborize primarily in the M6 layer (Ting et al., 2014) (Fig. 2A,B). In PlexA mutant pupae, Dm8 projections were disordered and extended into a wider region of the medulla, enabling them to contact the mis-projecting R7 axon terminals (Fig. 2C). To ascertain whether these contacts contained synaptic components, we expressed a tagged short version of the presynaptic protein Bruchpilot (Brp) in a subset of R7 cells (Fig. S1E,F). Pan-neuronal knockdown of PlexA using elav-GAL4 disrupted R7 targeting, but Brp puncta were still present in R7 terminals, suggesting that synapses formed despite the abnormal position of R7-Dm8 contacts (Fig. S1F). These knockdown flies survived to adulthood, implying that the late pupal lethality of null PlexA mutants is due to functions of PlexA in non-neuronal cells.

The effect of PlexA on Dm8 dendrite projections suggested that, rather than specifically guiding R7 photoreceptors, PlexA might play a more general role in establishing medulla neuron projection morphology. We therefore examined how loss of PlexA affected targeting by several classes of medulla neurons: medulla intrinsic (Mi) neurons, transmedullary (Tm) neurons that project through the medulla to the lobula complex, local medulla interneurons and serpentine medulla (Sm) neurons, which project together with tangential neurons in the M7 layer of the medulla but do not extend axons into the central brain (Fig. 2A) (Matsliah et al., 2023 preprint). We labelled Mi1 neurons, Tm5a, Tm5b and Tm5c neurons, Tm9 neurons, local interneurons in the M6 layer, and a subset of Sm neurons in the M7 layer with UAS-myrTomato driven by bshM-GAL4 (Han et al., 2020; Trush et al., 2019), GMR9D03-GAL4 (Douthit et al., 2021; Han et al., 2011), GMR24C08-GAL4 (Fisher et al., 2015), 30A-GAL4 (Chin et al., 2014) and GMR35A02-GAL4 (Jenett et al., 2012), respectively. Mi1 neuron projections are among the earliest to enter the medulla, and arborize in medulla layers M1, M5 and M9-10 (Fig. 2A). In PlexA mutants, Mi1 neurites lost their distinct arborizations in layer M5, instead sending projections throughout the medial medulla (Fig. 2D,E,N; Table S1). Tm5 and Tm9 have dendritic arbors in the medulla, and project their axons into the lobula (Fig. 2A). The dendrites of both cell types were less compactly stratified in PlexA mutants compared with wild type in the M3 layer for both neurons and in the M8 layer for Tm5 (Fig. 2F-I,N; Table S1). The neurites of 30A-GAL4-expressing interneurons appeared tightly apposed in the M6 layer in wild-type brains (Fig. 2A,J; Table S1). In PlexA mutants, these interneurons instead extended projections through many layers of the medulla (Fig. 2K,N; Table S1). Sm neurons send processes horizontally across the medulla in the M7 layer (Fig. 2A,L); in PlexA mutant flies, these projections were not confined to M7, but wandered extensively throughout the proximal medulla (Fig. 2L-N; Table S1). Finally, we used an anti-Sdk antibody to label the dendrites of T4 lobula plate neurons in layer M10 (Astigarraga et al., 2018), and observed that Sdk staining in this layer occupied a larger proportion of the width of the medulla in PlexA mutants than in controls (Fig. S2A-C). Loss of PlexA therefore disrupts the projection patterns of neurites from multiple types of neurons that project in many of the synaptic layers of the medulla.

To characterize how loss of PlexA affects the overall organization of the medulla, we stained PlexA mutant pupal brains 72 h after puparium formation (APF) for the postsynaptic protein Discs-large (Dlg). In wild-type optic lobes, Dlg showed a consistent pattern that varied in intensity between individual medulla layers, with notably high intensity in layer M2, in the proximal half of M3, and in M5, M9 and M10 (Chin et al., 2014) (Fig. 3A). In PlexA mutants, Dlg was not as tightly restricted to distinct layers as in wild-type pupae (Fig. 3B), and the peaks and valleys of Dlg intensity that characterize the wild-type medulla were flattened (Fig. 3C). The neuronal cadherin family member N-cadherin (Ncad) was present throughout much of the wild-type medulla neuropil, with three distinct regions of low Ncad intensity in layers M3, M5 and the proximal half of M7 (Fig. 3D,F). In PlexA mutants, Ncad staining was more uniform across the medulla (Fig. 3E,F). We quantified the difference in the patterns of Ncad and Dlg enrichment by measuring the sinuosity (the length of the measured intensity curve divided by the length of a straight line connecting the endpoints) of the Ncad and Dlg proximal-distal intensity curves (Fig. S2D). The sinuosity of both curves was significantly reduced in PlexA mutants (Fig. 3G). Most of the differences in intensity in PlexA mutants affected the medial medulla, with layers M1, M2 and M10 appearing relatively normal (Fig. 3C,F). The poorly defined layers in the PlexA mutant medulla may account for the misprojection of photoreceptor axons.

We next investigated in which cells PlexA was required for medulla lamination. Expressing PlexA RNAi in all neurons with elav-GAL4 resulted in shortened and disorganized photoreceptor axon projections (Fig. S1E,F) and diffuse medulla layering observed with anti-Ncad staining (Fig. S2E,F,H), indicating that PlexA acts in neurons. In contrast, expressing PlexA RNAi in all glia with repo-GAL4 did not significantly affect medulla layering (Fig. S2G,H). In the pupal medulla, PlexA protein is enriched in the developing M7 layer (Fig. 3H-K; Fig. S2I-L), where it colocalizes with the processes of medulla tangential neurons (Pecot et al., 2013). Medulla neurons arise from two neuronal progenitor populations: the main outer proliferation center (OPC) and the tips of the OPC (tOPC) (Bertet et al., 2014; Li et al., 2013), also referred to as the glial precursor cells (GPC) (Perez and Steller, 1996; Suzuki et al., 2016). Neuroblasts that form the main OPC express a temporal series of transcription factors beginning with Homothorax (Hth) and go on to generate most of the neurons in the medulla (Li et al., 2013; Suzuki et al., 2013). Neuroblasts in the tOPC, which express a distinct series of transcription factors beginning with Distal-less (Dll), give rise to medulla tangential neurons among other cell types (Bertet et al., 2014). To determine in which neuroblast lineage PlexA acts, we expressed UAS-Cas9 and PlexA sgRNAs with hth-GAL4 and Dll-GAL4 drivers to mutate PlexA in each population of neuroblasts by somatic CRISPR. Loss of PlexA in neurons of the hth-GAL4 lineage did not significantly affect photoreceptor axon targeting or Ncad staining (Fig. S2M,N,P). In contrast, loss of PlexA in neurons of the Dll-GAL4 lineage resulted in shortened and disorganized R7 and R8 axon projections, as well as less distinct Ncad layering (Fig. S2M,O,P), suggesting that PlexA is required in neurons derived from tOPC neuroblasts. The weaker phenotype compared with PlexA mutants could reflect mosaic rather than complete deletion of the PlexA gene, and/or a contribution of additional neurons to PlexA function. Although PlexA was enriched in the M7 layer during mid- and late pupal development, some PlexA protein was detected throughout the medulla neuropil (Fig. S2I-K), and could be clearly visualized using a Myc-tagged form of endogenous PlexA (Fig. 3I-K). Our data are consistent with a role for PlexA expressed by M7 tangential neurons formed from tOPC neuroblasts in the establishment of medulla layers, but these cells may not be the only source of PlexA.

Interactions between PlexA and the semaphorin family of proteins are important for axon guidance (Negishi et al., 2005; Tamagnone et al., 1999) as well as other developmental processes, such as epithelial remodeling after wound healing (Yoo et al., 2016) and collective cell migration (Stedden et al., 2019). Plexins and semaphorins interact through their highly conserved Sema domains (Janssen et al., 2010; Winberg et al., 1998). The bulk of the Sema domain of PlexA is encoded by an exon that is missing from splice isoforms RB and RC (Fig. 1C). To determine whether the Sema domain is necessary for the function of PlexA in medulla layer formation, we specifically mutated this exon in the PlexA gene using CRISPR (PlexAΔSema). Western blotting confirmed the loss of full-length PlexA and the presence of low levels of PlexAΔSema protein in mutant larval brains (Fig. S3C). In PlexAΔSema mutant pupae, photoreceptor axon targeting was disrupted and the medulla layers labeled by Ncad and Dlg were wider and more diffuse (Fig. S3A,B), similar to PlexA null mutants. This suggests that isoforms containing the Sema domain (RA and RF) are necessary to pattern the medulla layers.

Three Drosophila semaphorins – Sema1a, Sema1b and Sema5c – interact with PlexA (Stedden et al., 2019; Winberg et al., 1998) and are expressed in the pupal optic lobe throughout development (Kurmangaliyev et al., 2020; Ozel et al., 2021). Sema1a protein is present within the medulla as early as 12 h APF, but is excluded from the putative M7 layer (Pecot et al., 2013). To determine which semaphorins contribute to medulla layering, we stained Sema1a^P1 (Yu et al., 1998), Sema1b^KO (Wittes and Schüpbach, 2019) and Sema5c^K175 (Stedden et al., 2019) mutant pupal medullas for Dlg and Ncad (Fig. 4A,B; Fig. S3D,E). Sema1b and Sema5c mutants both showed normal medulla layers visualized with these markers (Fig. S3D,E). In Sema1a mutants these layers were less distinct, with a corresponding decrease in the sinuosity of Dlg and Ncad measurements (Fig. 4C-E). However, Ncad layers were affected primarily in the medial medulla (between the 20th and 70th percentiles of medulla height) (Fig. 4B′,D), and distal Dlg layers appeared better defined than in PlexA mutants (compare Fig. 4B,C,E with 3B,C,G). We also found that medulla neurons extended neurite projections across a broader extent of the medulla neuropil in Sema1a mutants than in controls (Fig. S3J-M, Table S1).

The failure of Sema1a^P1 mutants to fully phenocopy the layering defect observed in PlexA mutants suggested that Sema1a could be partially redundant with other semaphorin family members. Both Sema1a, Sema1b and Sema1a; Sema5c double mutants died before pupation, precluding the study of medulla layering. Therefore, we used RNAi knockdown in mutant backgrounds to remove different combinations of semaphorin proteins from the developing brain. Knockdown of Sema5c in a Sema1a mutant, or of Sema1a in a Sema1b mutant, did not result in additional layering defects compared with Sema1a alone (Fig. S3F,G), and Sema1b, Sema5c double mutants did not disrupt medulla layers visualized by Dlg or Ncad staining (Fig. S3H). These experiments thus did not reveal any redundancy within the semaphorin family that could explain the differences between PlexA and Sema1a mutant phenotypes. Recently, a null allele of Sema1a was generated using CRISPR/Cas9 (Sema1a^SK1, National Institute of Genetics). Very few Sema1a^SK1 homozygous or hemizygous mutants survived to the third larval instar, suggesting that complete loss of Sema1a is lethal, and that the Sema1a^P1 allele is not null. We obtained a single escaper Sema1a^SK1 hemizygous pupa, in which Ncad- and Dlg-stained layers were slightly less defined than in Sema1a^P1 mutants, but it did not appear as severe as PlexA (Fig. S3I).

Sema1a has been shown to autonomously control the targeting of lamina neuron projections to the outer medulla by transducing a repulsive activity of PlexA, the effect of which was examined by PlexA RNAi knockdown (Pecot et al., 2013). We confirmed that PlexA null mutants showed a similar defect in lamina neuron projections (Fig. 4F,G). To determine whether the changes in overall layering of the medulla in Sema1a mutants were a consequence of its function in lamina neurons, we used a lamina precursor driver to knock down expression of Sema1a and examined the effects on Ncad staining in the medial domain (20-70%) of the medulla. We found only a small effect on overall medulla layering (Fig. S4G,H), although the same RNAi line disrupted layering more severely when expressed in all neurons with elav-GAL4 or nSyb-GAL4 (Fig. S4A-F), suggesting that loss of Sema1a from medulla neurons contributes to the layering defects.

To determine which neurons require Sema1a autonomously for layering, we used published single-cell RNA sequencing data (Kurmangaliyev et al., 2020; Ozel et al., 2021) to identify cell types that strongly express Sema1a during early pupal stages. Among these cells, we chose to investigate Tm5 and Mi1 neurons because of the availability of early GAL4 drivers. Knockdown of Sema1a in Tm5 neurons (Douthit et al., 2021; Han et al., 2011), resulted in Tm5 projections that were less restricted in layers M3 and M8 than in controls (Fig. 4H,I,N; Table S1), consistent with less distinct Ncad patterning in the medial medulla (Fig. S4I,J). Knockdown of Sema1a in Mi1 neurons did not significantly affect their arborizations in layers M1 or M9-10, or the Ncad staining pattern, but Mi1 arborizations in the medial medulla (M3) became broader (Fig. 4J,K,N; Fig. S4M, Table S1). Knocking down Sema1a in Tm9, M6 and Sm neurons did not significantly affect either their projections or the Ncad staining pattern (Fig. 4L-N; Fig. S4K,L,O,P, Table S1). Although Sema1a is required autonomously in only a subset of the neurons we tested, additional neurons showed disrupted medulla arborization in Sema1a mutants (Table S1). Cells that do not require Sema1a autonomously may still rely on Sema1a as a ligand to direct their targeting, or respond to other cues from Sema1a-expressing cells.

Although PlexA was first described as a receptor for Sema1a, in some contexts their roles are reversed, with PlexA acting as a ligand for Sema1a to activate a downstream ‘reverse signaling’ pathway (Battistini and Tamagnone, 2016). If PlexA acts as a receptor, it should be autonomously required in the medulla neurons that are mistargeted in PlexA mutants. To test this model, we knocked down PlexA in these neurons and quantified the extent of their arborizations within the medulla. We found that, although pan-neuronal knockdown of PlexA results in a significantly disordered layering pattern (Fig. S2F,H) similar to the PlexA mutant, knockdown of PlexA in most neuronal subtypes did not affect their arborizations in the medulla (Fig. S5S-V, Table S1). One exception was PlexA knockdown in serpentine medulla neurons, which resulted in significant neurite targeting outside the serpentine layer, suggesting that PlexA acts cell autonomously to guide these neurons (Fig. S5O,P, Table S1).

The PlexA cytoplasmic domain is crucial for its function as a receptor; it contains a RasGAP domain and GTPase binding domain (Fig. 5A), and also interacts with other signaling components, such as Mical and the guanylyl cyclase Gyc76C (Chak and Kolodkin, 2014; Hung et al., 2010; Yu et al., 2010). To confirm that PlexA acts as a ligand to influence Sema1a-expressing neurons, we used CRISPR to insert a premature STOP codon 50 amino acids after the transmembrane domain of PlexA, generating a mutant allele lacking the cytoplasmic domain (PlexAΔcyto) (Fig. 5A; Fig. S5A) that would be unable to function as a receptor. An HA tag was added to the C terminus to allow the monitoring of protein production; western blotting confirmed the presence of PlexAΔcyto protein in the larval brain (Fig. S5B). HA staining of late pupal brains showed that PlexAΔcyto protein was present in neurites in the medulla, but was not as enriched in the M7 layer as full-length PlexA, suggesting that the cytoplasmic domain contributes to selective trafficking or degradation of the protein (Fig. S5C,D). Surprisingly, PlexAΔcyto flies could survive to adulthood, indicating that the receptor function of PlexA is not essential for viability (Fig. S5E).

To test the role of the cytoplasmic domain of PlexA in medulla layering, we stained PlexAΔcyto pupal brains for Dlg, Ncad and Sdk, as well as Choline acetyltransferase (ChAT) and Connectin (Con), a homophilic adhesion protein involved in axon guidance (Fig. 5B-G; Fig. S5C′,D′,F-L). The layers observed with these markers were more distinct than in PlexA null mutants (Fig. 5B-G; Fig. S5C,D,F-L). In addition, most R7 photoreceptors terminated correctly in the M6 layer (Fig. 5H,I), and M6, Mi1, Tm5 and Tm9 neuron projections were less affected in PlexAΔcyto mutants than in PlexA null mutants (Fig. 5J-N; Table S1). The dispersion of the M3 neurites of Tm5 in PlexAΔcyto mutants may be an indirect effect of loss of PlexA receptor function in other neurons, as the M8 neurites of the same cell were unaffected (Fig. S5Q, Table S1).

These data indicate that PlexA expressed at endogenous levels can mediate substantial layering without its cytoplasmic domain, making it likely that it acts primarily as a ligand rather than a receptor in this context. Serpentine medulla neuron projections remained significantly disordered in PlexAΔcyto mutant pupae (Fig. S5M,N,P, Table S1), supporting an autonomous receptor function for PlexA in these neurons.

In addition to establishing synaptic layers, the medulla neuropil undergoes substantial morphological changes during pupal development. In early pupal stages, transverse optical sections of the medulla neuropil stained for Ncad have a distinct wedge shape, which is narrower at the posterior than in the more-developed anterior region (Ngo et al., 2017) (Fig. 6A). As development proceeds, the medulla acquires an even height throughout when viewed in horizontal sections (Fig. 6B). After 48 h APF, the medulla increases in height as synaptic layers increase in resolution (Ngo et al., 2017) (Fig. 6C).

To characterize how loss of PlexA affects medulla morphogenesis, we stained early-, mid- and late-stage pupal brains for Ncad, and quantified the anterior-posterior width and proximal-distal height of the medulla neuropil in horizontal sections (Fig. 6A-F,M,N). PlexA mutant medullas displayed a less pronounced wedge shape at 24 h APF than controls (Fig. 6A,D), suggesting that PlexA establishes the correct neuropil shape at the onset of pupal development. Pupal medullas homozygous for either PlexA²² or PlexA^MB09499 showed significant decreases in both width and height at 72 h APF compared with controls, and the decreased width was already apparent at 48 h APF (Fig. 6B,C,E,F,M,N; Fig. S6A,B,G-I). The opposite effect was observed in frontal sections, in which the wild-type medulla forms an arc distal to the lobula and lobula plate neuropils (Fig. 6J). In PlexA mutants, the medulla increased in length along the dorsal-ventral axis, wrapping around the lobula complex in a semicircular shape (Fig. 6K,O). We measured the volume of the medulla neuropil using 3D optical stacks, and found no significant difference between control and PlexA mutant samples (Fig. S6M, Movies 1 and 2). There was also no significant difference in the number of ommatidia or of Tm9 neurons in PlexA mutant pupae compared with controls (Fig. S6N-P). Together, these data show that the defect in PlexA mutants is a change in the aspect ratio of the medulla neuropil rather than a reduction in its volume.

We next asked whether PlexA regulates medulla morphogenesis as a ligand or a receptor. Although loss of the PlexA cytoplasmic domain did not significantly affect the height of the medulla neuropil, consistent with its limited effect on the segregation of medulla layers, the width of horizontal medulla neuropil sections was reduced to a similar extent in PlexAΔcyto pupal brains as in PlexA null mutants, beginning at the same stage of development (Fig. 6G-I,M,N; Fig. S6I). In frontal cross-sections, PlexAΔcyto mutants showed an increase in medulla length along the dorsal-ventral axis (Fig. 6L,O), similar to PlexA null mutants, suggesting that PlexA acts as a receptor to mediate medulla morphogenesis. We also observed changes in the angle between the medulla and lamina in PlexA²² and PlexAΔcyto mutants (Fig. S6Q-S,U), consistent with a requirement for PlexA signaling during medulla rotation. Although Sema1a^P1 mutants also showed abnormal rotation (Fig. S6T,U), medulla width was not reduced to the same extent as in PlexA mutants (Fig. S6C,D,G,H), indicating that other ligands may regulate this function of PlexA. Knockdown of PlexA in neurons, but not glia, resulted in a partial phenocopy of the PlexA mutant morphology defect (Fig. S6E-H,J-L), implicating neurons as the source of PlexA that mediates neuropil shape. Thus, PlexA likely acts as a ligand for Sema1a to segregate synaptic layers in the medulla, but as a receptor – perhaps for a different ligand – to give the medulla neuropil its overall shape (Fig. 6P).

Transmembrane semaphorins have been implicated as receptors that activate a reverse signaling pathway in several vertebrate and invertebrate systems (Battistini and Tamagnone, 2016; Andermatt et al., 2014; Sun et al., 2015; Toyofuku et al., 2004; Yu et al., 2010). However, the evidence that plexins act as their ligands has largely relied on overexpression experiments (Hung et al., 2010; Yu et al., 2010). By using CRISPR mutagenesis to remove the cytoplasmic domain from the endogenous PlexA protein, we have shown conclusively that its effect on medulla layer formation is minor in comparison with the null PlexA mutant, consistent with a ligand function for PlexA in this process (Fig. 5B-G, 6P; Fig. S5I-L). Our data also point to Sema1a as the primary receptor for this function of PlexA (Fig. 4A-E; Fig. S3J-M, Table S1), although we cannot exclude a redundant contribution from other semaphorins.

In contrast, the cytoplasmic domain of PlexA is required for normal morphogenesis of the medulla neuropil (Fig. S6G-I,L-P), suggesting that this process relies on a canonical receptor function of PlexA. The ligand for PlexA in medulla morphogenesis remains to be identified, as Sema1a^P1 does not alter medulla neuropil shape to the same extent as PlexA²² (Fig. S6C,D,G,H). Sema1a might act redundantly with other semaphorins to direct medulla morphogenesis; the early lethality of combinations of Sema mutations prevented us from testing this hypothesis. Another possible ligand is the secreted protein Slit, which is required in medulla neurons and glia to regulate morphogenesis of the optic lobe neuropils and to establish boundaries between them (Caipo et al., 2020; Suzuki et al., 2016; Tayler et al., 2004). At the mouse midline, a C-terminal fragment of Slit has been shown to bind to plexin A1, inducing repulsive activity that is independent of Robo and neuropilin receptors (Delloye-Bourgeois et al., 2015).

The receptor function of PlexA is required in neurons, and controls the shape of the medulla neuropil without changing neuropil volume or overall cell number, consistent with the strong correlation between the number of olfactory sensory neurons and the volume of their corresponding glomerulus in both mice and flies (Bressel et al., 2016; Grabe et al., 2016). The shape and internal architecture of the C. elegans nerve ring neuropil depend on the ingress of a set of pioneer neurons into the putative neuropil region, providing an initial organizing cue to targeting neurons (Moyle et al., 2021). As PlexA acts early in development, while neurogenesis and targeting are occurring, it may similarly regulate neuropil shape through the positioning or extent of neurite ingress into the medulla neuropil. Such a role would be consistent with the effects of the cytoplasmic domain of PlexA on the actin cytoskeleton and integrin-mediated adhesion through its Mical and Rap effectors (Yang and Terman, 2013).

Surprisingly, homozygous mutants lacking the PlexA cytoplasmic domain are viable. Thus, although PlexA acts as a receptor in many processes in Drosophila, including motor axon guidance (Winberg et al., 1998), development of olfactory circuits (Sweeney et al., 2007), cell migration in the ovary (Stedden et al., 2019) and epithelial repair after wound healing (Yoo et al., 2016), its receptor function is not crucial for survival.

PlexA-Sema1a signaling has previously been reported to restrict lamina neuron projections to the distal medulla (Pecot et al., 2013). Our results extend this analysis to other neurons that project into the medulla, and show that loss of PlexA leads to broader and overlapping synaptic layers. PlexA has a similar effect on the segregation of laminae in the ellipsoid body, a neuropil substructure located in the central brain that is innervated by ring neurons (Hanesch et al., 1989; Xie et al., 2017). Sema1a appears to act as a PlexA receptor in R4m ring neurons to prevent them from forming synapses in the central ellipsoid body, and loss of Sema1a from these neurons in turn affects R3 neurons, which expand their arborizations into the outer ellipsoid body (Xie et al., 2017). A similar non-autonomous function might explain how PlexA and Sema1a pattern the medulla. Single-cell RNA sequencing data indicate that Sema1a expression levels vary in different types of medulla neurons (Kurmangaliyev et al., 2020; Ozel et al., 2021), suggesting that some neurons may rely more heavily on Sema1a signaling. Furthermore, we show that knockdown of Sema1a in Tm5 or Mi1 neurons disrupts their dendritic arborization patterns, whereas other cell types such as Tm9 neurons do not autonomously require Sema1a. We propose that a subset of medulla neurons directly responds to PlexA through Sema1a signaling, and they require Sema1a cell-autonomously for proper localization and patterning of their dendritic arbors (Fig. 6P). Other cells do not directly sense PlexA, but are disrupted in Sema1a or PlexA mutants because they are indirectly affected by neurons that autonomously require Sema1a or PlexA (Fig. 6P). Finally, serpentine neurons that project in the M7 layer, where PlexA levels are highest, require a cell-autonomous PlexA function mediated by the PlexA cytoplasmic domain, most likely as a receptor for Sema1a.

Although some molecular cues that establish synaptic lamination are themselves segregated into layers (Yamagata and Sanes, 2008, 2012), this does not seem to be the case for PlexA. PlexA protein is enriched in the M7 layer and its effects on layer formation appear weakest in the layers furthest from M7, but it is also present at lower levels throughout the medulla. Our somatic CRISPR experiments support a requirement for PlexA in tangential neuron precursors, but do not exclude a role for it in other neurons, as the knockout achieved by this method is unlikely to be complete. The widespread distribution of PlexA excludes a model in which PlexA is always a repulsive signal, as in this case Sema1a-expressing neurites would fail to enter the medulla neuropil. It is possible that different levels of PlexA have distinct attractive or repulsive effects, or that PlexA has a function in cis in Sema1a-expressing cells (Rozbesky et al., 2020; Sun et al., 2013), explaining its autonomous activity in some medulla neurons (Table S1). PlexA and Sema1a could also influence one another's location through localized binding or receptor-mediated endocytosis, leading to local differences in protein levels that might not be detectable by antibody staining but could contribute to the establishment of layer boundaries. Additionally, the dynamic nature of medulla layering (Ngo et al., 2017) means that a growing neurite may receive locational cues from cells that later change their position or gene expression, rendering the temporal as well as spatial patterns of protein distribution relevant for targeting.

Plexin-semaphorin signaling plays a variety of roles in layer-specific neurite arborization in the mouse retina and in retinal ganglion cell targeting (Matsuoka et al., 2011; Sun et al., 2015, 2013). Our findings provide further evidence that interactions between plexins and semaphorins regulate neuropil organization through multiple mechanisms, and strengthen the findings of structural and functional homology between the fly optic lobe and mouse retina (Sanes and Zipursky, 2010). Furthermore, the role of PlexA in medulla neuropil morphogenesis establishes a molecular connection between the gross morphology of a brain structure and its internal synaptic organization.

Fly stocks used were PlexA²² (this paper); PlexA^MB09499 (BDSC#61741); PlexAΔcyto-HA (this paper); PlexAΔSema (this paper); Sema1a^P1 (BDSC 11097); Sema1a^SK1 (NIG-Fly M2L-3127); Sema1b^KO (Wittes and Schüpbach, 2019); Sema5c^K175 (Stedden et al., 2019); Df(2 L)Exel7039 (BDSC 7810); UAS-CD8-GFP (Kyoto 108068); UAS-myrTomato (BDSC 32221); UAS-dcr2 (BDSC 24650); UAS-Cas9-P2 (from BDSC 54593); repo-GAL4 (BDSC 7415); nSyb-GAL4 (BDSC 51635); elav-GAL4 (BDSC 8765); hth-GAL4 (Wernet et al., 2003); Dll-GAL4 (BDSC 3038); 30A-GAL4 (BDSC 37534); GMR35A02-GAL4 (BDSC 49811); GMR24C08-GAL4 (BDSC 48050); GMR9D03-GAL4 (BDSC 48050); bshM-GAL4 (Trush et al., 2019; Han et al., 2020); ort^C2b-GAL4 (Ting et al., 2014); GMR9B08-GAL4 (BDSC 41369); Rh3-LexA (Mazzoni et al., 2008); LexAop::brpShort-mCherry (Mosca and Luo, 2014); Sema1a RNAi HMS01307 (BDSC 34320); Sema5c RNAi JF03372 (BDSC 29436); PlexA RNAi HM05221 (BDSC 30482); PlexA RNAi KK101499 (VDRC v107004); pCFD4-PlexA sgRNAs (this paper); w¹¹¹⁸ (BDSC 3605); and Bac-PlexA-Myc (Pecot et al., 2013).

The genotype used to generate large PlexA mutant clones in a Minute background was ey3.5-FLP; tub-GAL4, UAS-myrTomato; FRT80, Rh3/4-lacZ / FRT80, M(3)67C, tub-GAL80, Dp(4;3)RC051; PlexA^MB09499. repo-GAL4 was used to drive gene expression in glial cells; elav-GAL4 and nSyb-GAL4 were used to drive gene expression in neurons. hth-GAL4 and Dll-GAL4 were used to drive gene expression in medulla and tangential neuron precursors, respectively (Bertet et al., 2014). GMR35A02-GAL4 was used to label likely serpentine medulla neurons in the M7 layer of the pupal medulla (Matsliah et al., 2023 preprint). 30A-GAL4 was used to label M6 local medulla neurons in the pupa (Chin et al., 2014). GMR24C08-GAL4 was used to label Tm9 neurons (Chin et al., 2014). Dm8 neurons were labeled with ort^C2b-GAL4 (Ting et al., 2014). GMR9D03-GAL4 was used to label Tm5a, Tm5b and Tm5c neurons (Douthit et al., 2021; Han et al., 2011). bshM-GAL4 was used to label Mi1 neurons (Han et al., 2020; Trush et al., 2019). GMR9B08-GAL4 was used to drive gene expression in the precursors of lamina neurons (Pecot et al., 2013; Schwabe et al., 2014). Both male and female flies were analyzed, and w¹¹¹⁸ flies were used as a wild-type control throughout.

To generate the PlexA²² mutant allele, the PlexA sgRNA sequences CATTACTTCAGTACCGGTGG and GTTGACGCTTGTACACATGA were made by gene synthesis in pUC57 (GenScript) and cloned into pCFD4 (Port et al., 2014) by Gibson assembly. The pCFD4-null construct was integrated into the attP40 site at 25C6. These flies were crossed to nos-Cas9, and their progeny were crossed to Tb, RFP/ci^D flies and screened first for failure to complement PlexA^MB09499, and then (by PCR) for the expected deletion, using the primers GGCGTGGCTATTCGTTATTCG and TGCGACACAATTGGACAAGTG.

To generate the PlexAΔSema mutant, the sgRNA sequences ACTTTCCTTATTGTTTTTTC and TGTGTTCTAACGGTAAGTAT were made by gene synthesis in pUC57 (GenScript) then cloned into pCFD4 (Port et al., 2014) by Gibson assembly. The pCFD4-Sema construct was integrated into the attP40 site at 25C6. Injections and screening of transgenic flies were carried out by GenetiVision. After crossing to nos-Cas9 flies and screening by PCR, the deletion expected if the DNA had been cut at both sgRNA positions was not obtained. The lethal mutations were characterized by PCR using the primers CGTAGCATTTATTGCGCCG and CGTGGGAATTAACTTTAAAGGGC. We isolated one mutation that had a single base-pair insertion and one that had a two base-pair deletion immediately before the 3′ splice junction of the Sema domain-encoding exon. Both introduced a frameshift leading to a stop codon shortly thereafter, and would prevent transcripts containing this exon from producing a functional protein. Fig. S3A,B shows the single base-pair insertion.

The PlexAΔcyto mutant was generated via CRISPR mutagenesis by GenetiVision Corporation. A pCFD5 plasmid containing gRNA1 (Kanca et al., 2019) and a gRNA that targeted the PlexA gene immediately after the transmembrane domain (CTTCGCTCACAGGCGATCTTACTTC) was injected into nos-Cas9 flies together with a pUC57 plasmid containing a donor construct (synthesized by GENEWIZ/Azenta) flanked by gRNA1 target sites that would insert an HA tag and 3X STOP cassette. Successful insertion of the donor fragment was confirmed by PCR amplification and sequencing.

Pupal development stages were calculated from the white prepupal stage (0 h APF) at 25°C. Pupal brains were dissected in ice-cold phosphate-buffered saline (PBS), fixed in 4% paraformaldehyde in PBS for 30 min at room temperature and washed three times for 5 min in PBST (PBS+0.3% Triton X-100). Adult brains were dissected in ice-cold Schneider's medium and fixed in 4% paraformaldehyde in PBS for 30 min at room temperature. After washing, samples were blocked in PBST+10% donkey serum for 30 min before incubation with primary antibodies in PBST overnight (pupal brains) or for 72 h (adult brains) at 4°C. Samples were washed in PBST three times for 20 min each and incubated in secondary antibodies for 4 h at room temperature or overnight at 4°C for pupal samples, or for 40-45 h at 4°C for adult samples. Samples were washed in PBS three times for 20 min each and mounted in SlowFade Gold AntiFade reagent (Invitrogen) on bridge slides.

For cryosections, adult heads were dissected in ice-cold 0.1 M sodium phosphate buffer (PB) (pH 7.4), fixed in 4% formaldehyde in PB for 4 h at 4°C and washed in PB. Heads were then submerged in a sucrose gradient (5%, 10% and 20%) for 20 min at each concentration, and left in 25% sucrose overnight at 4°C for cryoprotection. Heads were embedded in OCT tissue freezing medium and frozen in dry ice/ethanol, and 12 μm sections were cut on a cryostat. Sections were post-fixed in 0.5% formaldehyde in PB for 30 min at room temperature and washed three times in PB with 0.1% Triton (PBT) before incubation with primary antibodies overnight at 4°C. Sections were washed four times for 20 min with PBT and incubated with secondary antibodies for 2 h at room temperature. Sections were washed again four times for 20 min before mounting in Fluoromount-G (Southern Biotech).

Pupal retinas were dissected in PBS and fixed in 4% formaldehyde in 0.1 M PIPES (pH 7.0), 2 mM EGTA and 1 mM MgSO4 for 30 min. Retinas were washed for 15 min in PBS/0.2% Triton X-100 (PBT) and incubated overnight at 4°C with primary antibodies in 10% donkey serum in PBT. After three 5 min washes in PBT, they were incubated with secondary antibody in 10% donkey serum in PBT for 2-3 h at 4°C. After incubation with secondary antibody, retinas were washed again in PBT and mounted on glass slides in 80% glycerol in PBS.

Frontal and horizontal serial optical sections of whole-mount brain samples were acquired with 0.896 µm optical slices and 0.3 µm z-step intervals with a Leica SP8 laser scanning confocal microscope using a HC PL APO CS2 63×/1.40 oil objective, or a Zeiss LSM800 using an EC Plan-Neofluar 40×/1.30 oil DIC objective. Volumetric measurements were taken from z-stacks spanning the entire height of the optic lobe that were obtained with 1.272 µm optical slices and 0.6 µm z-step intervals using a Leica SP8 laser scanning confocal microscope with a HC PL APO CS2 40×/1.10 water objective. Images of the pupal retina were acquired on a Leica SP8 laser scanning confocal microscope using a HC PL FLUOTAR 10×/0.3 dry objective. Images were quantified in ImageJ/FIJI, and processed in ImageJ/FIJI and Adobe Photoshop. 3D reconstruction of confocal images was performed with FluoRender (Wan et al., 2012).

Primary antibodies used were mouse anti-Chp [1:50; Developmental Studies Hybridoma Bank (DSHB), 24B10], chicken anti-GFP (1:400; Invitrogen, A10262), rat anti-HA (1:50; Roche, 3F10), mouse anti-HA (1:200; BioLegend, 901513), rat anti-Ncad (1:50; DSHB, Ex#8), rat anti-Elav (1:100; DSHB, 7E8A10), rabbit anti-dsRed (1:500; Takara Bio, 632496), guinea pig anti-Sdk (1:200) (Astigarraga et al., 2018), mouse anti-ChAT (1:50; DSHB, 4B1), mouse anti-Con (1:50; DSHB, C1.427), mouse anti-Dlg (1:20; DSHB, 4F3), rabbit anti-PlexA (1:500) (Sweeney et al., 2007) or rabbit anti-Myc (1:1000; Abcam, ab9106). Secondary antibodies used were coupled to Alexa Fluor 488 (1:200; Invitrogen, chicken A11039, mouse A21202, rat A21208, rabbit A21206), Cy3 (1:200; Jackson ImmunoResearch, mouse 715-165-151, rat 712-165-153, rabbit 711-165-152), Cy5 (1:200; Jackson ImmunoResearch, mouse 715-175-151; rat 712-175-153; Invitrogen, rabbit A10523) or Alexa Fluor 647 (1:200; Jackson ImmunoResearch, guinea pig 706-605-148).

To extract proteins, L3 larval brains were dissected and homogenized in lysis buffer [50 mM Tris-HCl (pH 8), 150 mM NaCl, 1% Triton X-100, 0.5% deoxycholate, 0.1% SDS, 5 mM NaF, 1 mM Na₃VO₄ (pH 9.96), 5 mM EDTA, Complete Protease Inhibitor Cocktail (Roche)]. Samples were mixed with 4× Laemmli buffer [4% SDS, 20% glycerol, 120 mM Tris-Cl (pH 6.8), 0.02% Bromophenol Blue, 10% beta-mercaptoethanol] and heated at 95°C for 3 min before loading onto an SDS-PAGE gel. Gels were run first at 85 volts for 15 min, then 150 volts for the remainder of the time and transferred onto 0.45 µm nitrocellulose membranes (BioRad) for 2.5 h at 75 volts. Membranes were washed for 5 min in TBS and blocked in 5% low-fat milk in TBST [20 mM Tris (pH 7.6), 136 mM NaCl and 0.2% Tween-20] solution for 1 h. Membranes were incubated overnight with primary antibody in TBST with 5% milk at 4°C, washed three times for 10 min in TBST, and incubated in horseradish peroxidase-conjugated secondary antibodies (1:2000, Jackson ImmunoResearch) at room temperature in TBST with 5% milk for 2 h. Membranes were washed three times for 10 min in TBST and once for 10 min in TBS. Blots were developed with enhanced chemiluminescence (Thermo SuperSignal WestPico). Primary antibodies used were rat anti-Elav (1:2000, DSHB 7E8A10), rat anti-HA (1:2000, Roche 3F10), mouse anti-β-tubulin (1:10,000; Sigma T4026) and rabbit anti-PlexA (1:2000) (Sweeney et al., 2007).

All quantifications were carried out on maximum projections of 10 confocal sections (3 µm thick total stack). To quantify medulla layering in brains stained for Ncad, Dlg, Sdk, ChAT and Con, as well as neurite outgrowth in medulla neurons, the Straighten function in ImageJ/FIJI was used to straighten a region of the medulla representing greater than 50% of the overall neuropil. Image intensity of neuropil markers or neurite projections was measured over that region of interest using Plot Profile in FIJI, and the data array was resampled to include 300 data points, normalizing the medulla height of each sample. Intensity values were normalized by dividing all values in a sample by the highest value in that sample. Mean intensity profiles were plotted as a function of relative medulla height. The % neurite outgrowth was calculated by measuring the width of the corresponding intensity peak at 50% height, and dividing by the medulla height; measurements of each condition were averaged together. Sinuosity scores were calculated by measuring the length of each mean intensity profile line in ImageJ/FIJI, and dividing that value by the Euclidean distance between the endpoints of that line.

Measurements of overall medulla morphology were taken in ImageJ/FIJI, by measuring the length of a user-defined line from the distal-most anterior to posterior corners (medulla width), distal to proximal extent at the center-most position (medulla height) and along the distal-most neuropil boundary (medulla length). Medulla rotation was calculated by measuring the angles of user-defined lines from the distal-most anterior to posterior corners of the medulla and lamina relative to a horizontal line, and determining the difference between these two angles. To calculate medulla volume, images were first cropped around the medulla and background signal was removed using Otsu's thresholding, and voxels counted with the Voxel Counter plug-in. The thresholded volume (thresholded voxels×voxel size) is reported. To quantify the number of ommatidia, z-stacks were projected such that the entire retinal tissue was visible, and ommatidia were counted manually with the assistance of the Cell Counter plug-in in ImageJ/FIJI. Tm9 cell bodies were counted manually in ImageJ/FIJI with the Cell Counter plug-in from z-stacks spanning the entire optic lobe.

Statistical analysis was carried out using an unpaired t-test and one-way ANOVA with Tukey's multiple comparisons test or Dunnett's multiple comparisons test in GraphPad Prism. Sample size was not predetermined based on statistical measures.

We thank Claude Desplan, Isabel Holguera, Sally Horne-Badovinac, Alex Kolodkin, Chi-Hon Lee, Liqun Luo, Makoto Sato, Takashi Suzuki, Julia Wittes, Larry Zipursky, the Bloomington Drosophila Stock Center (NIH P40OD018537), the Vienna Drosophila Resource Center, the Kyoto Stock Center, the National Institute for Genetics and the Developmental Studies Hybridoma Bank for fly stocks and reagents, and FlyBase for essential information. We are grateful to DanQing He, Ariel Hairston and Dhaval Gandhi for expert technical assistance, to Michael Cammer for helpful discussions and assistance with image analysis, and to the NYU Microscopy Core for technical support. The NYU Microscopy Core (RRID: SCR_01934) is partially supported by the Cancer Center Support Grant P30CA016087 at the Laura and Isaac Perlmutter Cancer Center; instruments are supported by NIH grant S10 RR023708. The manuscript was improved by the critical comments of Neha Ghosh, Hongsu Wang, Chris Doe, Katherine Nagel and Niels Ringstad.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.202237.reviewer-comments.pdf