A defining characteristic of neuronal cell type is the growth of axons and dendrites into specific layers and columns of the brain. Although differences in cell surface receptors and adhesion molecules are known to cause differences in synaptic specificity, differences in downstream signaling mechanisms that determine cell type-appropriate targeting patterns are unknown. Using a forward genetic screen in Drosophila, we identify the GTPase effector Genghis khan (Gek) as playing a crucial role in the ability of a subset of photoreceptor (R cell) axons to innervate appropriate target columns. In particular, single-cell mosaic analyses demonstrate that R cell growth cones lacking Gek function grow to the appropriate ganglion, but frequently fail to innervate the correct target column. Further studies reveal that R cell axons lacking the activity of the small GTPase Cdc42 display similar defects, providing evidence that these proteins regulate a common set of processes. Gek is expressed in all R cells, and a detailed structure-function analysis reveals a set of regulatory domains with activities that restrict Gek function to the growth cone. Although Gek does not normally regulate layer-specific targeting, ectopic expression of Gek is sufficient to alter the targeting choices made by another R cell type, the targeting of which is normally Gek independent. Thus, specific regulation of cytoskeletal responses to targeting cues is necessary for cell type-appropriate synaptic specificity.
Sequences of genetically programmed developmental decisions play an important role in determining the complex architecture of the adult brain. At the level of single neurons, these decisions produce cellular morphologies that are characteristic of each cell type and reflect specific guidance and targeting choices made by both axons and dendrites. Although many molecules have been identified that affect synaptic specificity (Sanes and Yamagata, 2009), the molecular mechanisms that allow closely related cell types to make distinct targeting decisions in vivo are almost completely unknown. Here, we examine these mechanisms in the Drosophila visual system, and demonstrate that cell type-specific utilization of a cytoskeletal regulatory pathway plays a crucial role in the targeting choices made by a subset of R cells.
The visual systems of many animals are characterized by a broad organization of axons and dendrites into columns and layers (Sanes and Zipursky, 2010). Columns are organized into retinotopic arrays to allow the brain to process, in parallel, information from different points of visual space. Within each column, layers process different qualities of light and reflect different combinations of pre- and post-synaptic inputs. Closely related cell types make different choices with respect to these architectural features. How axons can be genetically programmed to innervate both the correct column and the appropriate layer, making cell type-specific targeting decisions, remains an open question.
Photoreceptor axons in Drosophila display both columnar and layer-specific targeting. The Drosophila visual system comprises the compound eye and four optic ganglia: the lamina, the medulla, the lobula and the lobula plate (Meinertzhagen and Hanson, 1993). Each facet of the eye, or ommatidium, contains eight R cells, which can be divided into three distinct types. R1-R6 cells, the outer photoreceptors in each ommatidium, project their axons into the lamina. Each of these axons innervates a specific column of post-synaptic target neurons, reconstructing a topographic map of the visual world in the brain. By contrast, the inner R7 and R8 cells project axons that terminate in two distinct layers in the medulla, making synaptic connections across multiple layers (Takemura et al., 2008). Cell ablation studies and genetic manipulations demonstrate that the targeting of R1-R6 axons to specific targets is both genetically hard-wired and dependent on interactions amongst afferent axons, rather than on specific interactions between afferents and their particular targets (Clandinin and Zipursky, 2000; Hiesinger et al., 2006). The layer-specific targeting of R7 and R8 cells, however, develops in a precise temporal sequence and is dependent upon specific interactions with presumed post-synaptic cells and intermediate targets (Senti et al., 2003; Shinza-Kameda et al., 2006; Ting et al., 2005; Tomasi et al., 2008). Thus, the cellular mechanisms that underlie columnar versus layer-specific targeting in this system are at least partially distinct.
Both histological and behavioral screens have identified many genes involved in the targeting of R1-R6 axons to appropriate columns within the lamina and in the targeting of R7 and R8 axons to appropriate medulla layers (Berger et al., 2008; Mast et al., 2006; Ting and Lee, 2007). For example, the classical cadherin N-cadherin mediates interactions between pre- and post-synaptic cells that are required for the targeting choices made by R1-R6 cells, and by R7 cells (Ting et al., 2005; Lee et al., 2001; Prakash et al., 2005; Yonekura et al., 2007). The receptor tyrosine phosphatases Lar and Ptp69D, as well as the Lar-interacting protein Liprin-α, stabilize these connections (Newsome et al., 2000a; Clandinin et al., 2001; Maurel-Zaffran et al., 2001; Choe et al., 2006; Hofmeyer et al., 2006; Hofmeyer et al., 2009; Prakash et al., 2009). Dynamic regulation of the expression of N-cadherin and Lar is also crucial to the layer-specific targeting decisions of R7 and R8 (Petrovic and Hummel, 2008). The non-classical cadherin Flamingo (Starry night – FlyBase) is required for appropriate topographic mapping, the layer-specific targeting of R8, and the columnar targeting choices made by R1-R6 (Chen and Clandinin, 2008; Lee et al., 2003; Senti et al., 2003). Intriguingly, for all genes where mutant phenotypes have been assessed in R1-R6, R7 and R8 photoreceptors, both columnar targeting by R1-R6, as well as layer-specific targeting by R7 and/or R8, were affected, even though the selection criteria under which they were initially identified were specific to only one of these phenotypes. Thus, these two different types of targeting decisions clearly use many of the same molecular components. However, it remains unclear whether additional components are required specifically for either the columnar targeting decisions of R1-R6 or the layer-specific targeting of R7 and R8.
Rho family GTPases regulate many stages of neuronal growth, including growth cone motility, axonal migration and dendritic spine morphogenesis (Govek et al., 2005). In the Drosophila nervous system, these GTPases regulate axon growth and guidance (Luo et al., 1996; Ng and Luo, 2004; Ng et al., 2002; Scott et al., 2003). In the Drosophila visual system, Rac is an upstream activator of the Dock-Pak signaling pathway, affecting R cell axon targeting to the lamina and the medulla (Hakeda-Suzuki et al., 2001; Hing et al., 1999; Newsome et al., 2000b). However, our understanding of the links between small GTPases and specific axonal targeting decisions remains incomplete, and only a limited set of GTPase effector proteins have been identified (Koh et al., 2006). One of these, Genghis khan (Gek), the Drosophila homolog of mammalian Cdc42-binding protein kinase alpha [CDC42BPα (or MRCKα); myotonic dystrophy-related Cdc42-binding kinase (MRCK-1) in C. elegans], was originally identified using biochemical approaches to identify effectors of the small GTPase Cdc42 (Leung et al., 1998; Luo et al., 1997). These studies demonstrated that Gek binds specifically to activated Cdc42, that this interaction increases the kinase activity of Gek in vitro, and that Gek activity alters actin cytoskeletal organization (Luo et al., 1997). Finally, genetic studies of asymmetric cell division and epithelial morphogenesis in C. elegans have demonstrated that MRCK-1 functions downstream of Cdc42 in vivo (Gally et al., 2009; Kumfer et al., 2010). However, the function of Gek in neurons is unknown. Here, we examine the role of Gek in controlling axon outgrowth and target selection in all subtypes of R cells in the Drosophila eye.
MATERIALS AND METHODS
Animal husbandry and genetics
Flies were maintained at 25°C on standard medium. Chemical mutagenesis was performed using ethylmethanesulfonate (EMS) under standard conditions (Clandinin et al., 2001). Eye-specific mosaic animals were generated using the FLP/FRT system, in which the expression of FLP is controlled by a fragment of the eyeless promoter in eyFLP or ey3.5FLP (Newsome et al., 2000a; Chotard et al., 2005). We used R11, a recessive cell-lethal mutation, to reduce the size of the twin spot (Chotard et al., 2005). MARCM studies were performed by expressing Gal4 under the control of the pan-neural elav promoter, and FLP was expressed under the control of the heat-shock promoter, as previously described (Prakash et al., 2005). The following alleles were used: gekNP5192 (this work), gekomb1080 (this work), flamingoe59, N-cadherinΔ14, Lar2127 and Cdc424 (Prakash et al., 2005; Maurel-Zaffran et al., 2001; Fehon et al., 1997; Usui et al., 1999).
A gek cDNA was obtained from EST clone LD24220 (Berkeley Drosophila Genome Project). Domain deletions were generated by PCR. All constructs were expressed using pUAST (Brand and Perrimon, 1993).
Generation of a Gek antibody
Anti-Gek antiserum was obtained for a rabbit immunized with a 6-His fusion protein containing 406 amino acids (433-838) of the coiled-coil domain region of Gek. The antiserum was purified against nitrocellulose containing the full-length protein and used at a dilution of 1:25.
Larval, pupal and adult brains were dissected and labeled as previously described (Clandinin et al., 2001). For structure-function studies of Gek, two independent insertions of each construct were scored for all constructs except Gek-ΔCC, where only a single insert was obtained. Between 10 and 37 larval brains of each genotype were examined. The following antibodies were used: mouse anti-Chaoptin mAb24B10 [1:25; Developmental Studies Hybridoma Bank (DSHB)], mouse anti-Repo (1:100; DSHB), goat anti-HRP FITC (1:100; Jackson), rat anti-mouse CD8 (1:100; Invitrogen), rabbit anti-β-galactosidase (1:100; Kappel), goat anti-mouse IgG Alexa Fluor 594 (1:200; Invitrogen), goat anti-rabbit IgG Alexa Fluor 488 (1:200; Invitrogen) and goat anti-rat IgG Alexa Fluor 488 (1:200; Invitrogen). The phospho-MRLC antibody was obtained from Cell Signaling Technology (Beverly, MA, USA) and used at 1:10.
A forward genetic screen identifies a gene specifically required for R1-R6 axon targeting
To identify mutations affecting photoreceptor axon targeting, we undertook a forward genetic screen based on a behavioral assay in which we assessed the ability of adult flies to see motion (Clandinin et al., 2001). In this approach, eye-specific somatic mosaic flies were made homozygous for EMS mutagenized chromosomes, while the rest of the fly remained heterozygous (and thus phenotypically wild type) (Newsome et al., 2000a; Stowers and Schwarz, 1999). These animals were then screened using an optomotor behavioral assay. Screening ∼15,000 mutagenized copies of the right arm of chromosome 2 identified 215 optomotor defective (omb) lines. We then examined the ability of mutant R cell axons to reach their post-synaptic targets in each of these lines, and identified a single mutation, designated omb1080, that caused specific defects in target selection.
To investigate the targeting of photoreceptor subsets in adult animals, we first examined the columnar targeting of R1-R6 axons within the lamina using the R cell-specific antigen Chaoptin. In control animals, R1-R6 cells choose post-synaptic targets that are arranged in a characteristic pattern, forming a reiterated array of barrel-like structures designated cartridges (Fig. 1A-C). In most of the target field, each cartridge contains exactly six R1-R6 terminals, reflecting a remarkably precise pattern of target selection. By contrast, in animals in which R cells were homozygous for omb1080, or bore omb1080 in trans to a deficiency, the overall array of cartridges was highly disorganized, with individual cartridges displaying variable numbers of axon terminals (Fig. 1D-F).
To examine the ganglion- and layer-specific targeting of all R cell axons we labeled R cells in eye-specific somatic mosaic animals using either the R1-R6-specific marker Rh1-lacZ, the R7-specific marker Rh4-lacZ, or the R8-specific marker Rh6-lacZ (Fig. 1G-N). R1-R6 axons homozygous for omb1080 almost invariably targeted to the lamina, with only rare fibers extending into the medulla (Fig. 1H,I). Thus, ganglion-specific targeting of R1-R6 axons was only subtly affected by omb1080. More strikingly, layer-specific targeting of R7 and R8 axons was virtually unaffected (Fig. 1I,J,M,N). In particular, 96% (n=432) of mutant R7 axons and 99% (n=503) of mutant R8 axons innervated the correct target layers in the medulla, as compared with 100% of control R7 and R8 axons (n=396 and n=580, respectively). Thus, omb1080 acts cell type specifically, affecting only the ability of R1-R6 axons to choose appropriate targets within the lamina. As the optomotor response is specifically dependent upon the function of R1-R6 cells, these defects are sufficient to explain the behavioral phenotype of omb1080.
omb1080 affects the serine-threonine kinase genghis khan
To identify the gene affected by omb1080 we used single nucleotide polymorphism (SNP) and deficiency mapping (supplementary material Fig. S1). As omb1080 was homozygous viable, these studies used direct histological examination of the targeting phenotype, and revealed that omb1080 corresponds to a tryptophan-to-stop amino acid change at position 374 in the genghis khan (gek) gene (Luo et al., 1997) (supplementary material Fig. S1). A second allele, gekNP5192, was subsequently obtained and consists of a Gal4 insert (in inverse orientation) just before the start codon (supplementary material Fig. S1). Both of these alleles were homozygous viable, caused highly penetrant R cell targeting phenotypes, and corresponded to strong reduction-of-function alleles (supplementary material Fig. S2; data not shown).
The predicted structure of Gek includes a serine/threonine kinase (STK) catalytic domain, a coiled-coil (CC) domain, a cysteine-rich region (CR), a pleckstrin homology (PH) domain, a citron homology (CH) domain and a Cdc42/Rac interactive binding (CRIB) domain (Luo et al., 1997) (supplementary material Fig. S1). This protein is highly similar in domain organization and primary sequence to the human CDC42BPα and CDC42BPβ (MRCKα and MRCKβ) isoforms (Luo et al., 1997) (supplementary material Fig. S1).
Gek is expressed on R cell axons and growth cones
To determine where Gek protein is expressed, we generated an affinity-purified polyclonal antibody directed against a bacterially expressed fragment of the coiled-coil domain. Using this antiserum, we examined Gek expression in the developing visual system at the third instar larval stage, at an early stage of pupal development [16 hours after puparium formation (APF)] when R7 and R8 axons project towards their intermediate targets, and at the mid-pupal stage when R1-R6 axons are extending to their targets (30 hours APF). In wild-type animals at the third larval stage, Gek expression was observed in R cell axons extending to the lamina, in growth cones within the lamina plexus, as well as within the medulla neuropil (Fig. 2A-D). Weak staining was seen in the R cell bodies in the eye disc (data not shown). No distinct staining pattern was detected in animals homozygous for gekNP5192, and photoreceptor clones homozygous for gekNP5192 lost Gek expression within the clone, demonstrating that the immunoreactivity we observed reflected endogenous Gek protein (Fig. 2E-H; data not shown). Early in pupal development, Gek was strongly expressed in R cell axons in both the lamina and the medulla (Fig. 2I-L). Similarly, at 30 hours APF, Gek was highly expressed on R1-R6 growth cones within the lamina plexus, and was also expressed in many other neurons in other optic ganglia (Fig. 2M-P). Thus, Gek is expressed in R cell growth cones as target selection occurs.
Loss of gek function disrupts R cell growth cone targeting to the lamina
To determine when Gek acts during R cell axon targeting, we first examined third instar larvae. In wild-type animals, R cell axon fascicles project through the optic stalk, forming a retinotopic array (Fig. 3A,B). At this stage, R1-R6 axons stop in the first optic ganglion, the lamina, whereas R7 and R8 axons project through the lamina plexus to terminate in the medulla. Within the lamina, R cell axon fascicles above and below the plexus are evenly spaced, and within the plexus R1-R6 cell growth cones form a uniform band of staining. By contrast, in transheterozygous animals bearing either gekNP5192 or gekomb1080 in trans to a small deficiency (Fig. 3C-E; data not shown), R cell axon fascicles above the lamina plexus were occasionally clumped and the lamina plexus appeared broken and varied in thickness. This phenotype was observed in all brains examined. Additionally, we occasionally observed bundles of axons extending into the medulla that were thicker than normal. Animals homozygous for either gekomb1080 or gekNP5192, or transheterozygous for both alleles, displayed indistinguishable phenotypes (data not shown). Similar, but weaker, phenotypes were also observed in somatic mosaic animals in which only R cells were made homozygous mutant using ey3.5FLP (Fig. 3E). These weaker phenotypes are consistent with Gek playing a specific role in R cell targeting during pupal development (see below), but might also reflect some perdurance of Gek protein in these mosaic clones.
As thickened bundles of axons extending into the medulla can be indicative of R1-R6 cell axons failing to stop in the lamina plexus, we directly labeled R2-R5 axons using ro-lacZtau (Callahan and Thomas, 1994; Heberlein and Rubin, 1990; Garrity et al., 1999). Consistent with previous observations, in control flies and gek heterozygotes bearing this construct, only rare R2-R5 cells extended into the medulla (Fig. 3F-I). In gek transheterozygous animals, we observed slightly increased numbers of R2-R5 cell axons projecting into the medulla (Fig. 3J; data not shown). However, overall, R2-R5 axons extending through into the medulla were rare, even in the most severely affected brains, indicating that virtually all R1-R6 axons stopped in the lamina in gek mutant animals.
To exclude the possibility that Gek indirectly affects targeting by regulating cell fate decisions, we stained mutant eye discs for the R7-specific Prospero protein or the R8-specific protein Senseless (Kauffmann et al., 1996; Frankfort et al., 2001). No transformations of R1-R6 neurons into R7 or R8 neurons were observed (supplementary material Fig. S3). We also examined the organization of lamina neurons in mutant flies using the pan-neural marker Elav, and found no differences between control and mutant animals (supplementary material Fig. S4). Finally, we assessed the organization of glial cells using Repo. In gek homozygous mutant animals, glia migrated normally into the lamina and were only subtly disorganized (supplementary material Fig. S4), which is likely to be a secondary consequence of the clumping of R1-R6 axon terminals in the lamina plexus. Thus, Gek does not cause errors in R cell axon targeting through indirect effects on target neuron or glial development.
Gek kinase activity is required for Gek function in R cell axons
To examine the function of each Gek domain, we attempted to rescue the gek mutant phenotype seen at the third larval stage using UAS-Gek constructs bearing either deletions of, or alterations in, individual domains (Fig. 4A). As controls, we expressed a full-length gek transgene (GekFL) specifically in the eye, under the control of the glass promoter, in both wild-type animals and in animals lacking Gek activity. We observed that overexpression of full-length Gek had no effect on axon targeting in wild-type animals, and that retinal expression was sufficient to strongly rescue the mutant phenotype (Fig. 4B,C; n=11/11 for wild type; n=23/28 indistinguishable from wild type in the Gek rescue).
We next examined whether the kinase domain was required for Gek function using two transgenes: Gek-ΔSTK, in which the kinase domain had been deleted; and Gek-K129A, in which the kinase domain had been inactivated using a point mutation demonstrated to eliminate kinase activity in vitro (Luo et al., 1997). Consistent with this domain being essential for Gek function, we observed little rescue of the R cell targeting phenotype when either of these transgenes was expressed in a gek mutant background (Fig. 4D; data not shown; 2/31 and 2/20 brains appeared indistinguishable from wild type for Gek-ΔSTK and Gek-K129A, respectively). Indeed, animals overexpressing the Gek-K129A transgene frequently displayed a severe phenotype, suggesting that this mutation might disrupt additional aspects of targeting. By contrast, gek transgenes that either deleted the Cdc42-binding domain (Gek-ΔCRIB), or which bore point mutations in the CRIB domain that prevent binding to activated Cdc42 in vitro (Gek-ISP) (Luo et al., 1997), fully rescued the gek mutant phenotype (Fig. 4E,F; n=17/18 and 21/27 indistinguishable from wild type for Gek-ΔCRIB and Gek-ISP, respectively). Thus, at least under conditions in which Gek protein is likely to be expressed at high levels, direct interactions between Cdc42 and Gek are not essential to Gek function. Similarly, expression of a Gek form that lacks the cysteine-rich domain, Gek-ΔCR, was also capable of fully rescuing the phenotype (Fig. 4G; n=18/22 indistinguishable from wild type). In vitro studies of mammalian CDC42BP demonstrate that this domain binds phorbol ester (Tan et al., 2001a). However, our data demonstrate that this interaction is not essential to Gek function.
Deletion of a subset of Gek domains causes gain-of-function phenotypes
We observed a very different phenotype when we expressed gek transgenes that lacked the coiled-coil domain (Gek-ΔCC), the pleckstrin homology domain (Gek-ΔPH) or the citron homology domain (Gek-ΔCH) (Fig. 4H-J). Expression of gek transgenes bearing any of these deletions in a mutant background caused early termination of many R cell growth cones, with large clumps appearing in the optic stalk and the distal lamina (10/11 Gek-ΔCC, 9/9 Gek-ΔPH and 21/22 Gek-ΔCH brains). These transgenes do not simply activate Gek. If this were the case, increasing Gek activity by restoring the wild-type gek locus should enhance these phenotypes. However, we observed that R cell axon targeting was normal when any of these three transgenes was expressed in wild-type backgrounds (Fig. 4K-M; n=16/16 Gek-ΔCC, 7/7 Gek-ΔPH and 4/4 Gek-ΔCH brains). Thus, deleting these domains did not simply constitutively activate Gek, nor are they strictly neomorphic (since neomorphic mutations cannot be competed by the presence of wild-type protein).
Gek is required cell-autonomously for R1-R6 cell axons to choose appropriate targets
To examine gek mutant phenotypes in individual growth cones, we used mosaic analysis with a repressible cell marker (MARCM) (Lee and Luo, 2001). To generate sparse somatic mosaic animals, we expressed FLP recombinase under the control of the heat-shock promoter and generated single R cells that were homozygous for either a control chromosome or a gek mutant chromosome in an otherwise heterozygous animal. These cells were labeled with mouse (m) CD8GFP under the control of the pan-neuronal promoter elavGal4, and thus their projections could be traced from the retina into the brain during mid-pupal development (Fig. 5). As the cell bodies of R1-R6 have characteristic positions and morphologies, each could be uniquely identified (Fig. 5A,A′,G,G′,M,M′,S,S′). Each R cell subtype chooses a single columnar target located in an invariant position with respect to the ommatidial axon bundle (Fig. 5B′,H′,N′,T′). When R cells are made homozygous for a control chromosome, they invariably made subtype-appropriate projections to the correct target (n=72/72 cells; Fig. 5B,H,N,T; supplementary material Fig. S5). Photoreceptors homozygous for either gekNP5192 or gekomb1080 invariably extended out of the retina into the brain, fasciculated normally, and stopped in the lamina (data not shown). However, once within the lamina, mutant photoreceptors often made errors in columnar targeting, making projections to one or more incorrect target cartridges in addition to the correct target, or failing to extend at all (Fig. 5C,D,I,J,O,P,U,V; supplementary material Fig. S5). In particular, for gekomb1080, only 35% (n=18/51) of R cells made the correct extension, whereas 35% (n=18/51) mistargeted and 30% (n=29/51) failed to extend (supplementary material Fig. S5). Targeting errors were observed in all R1-R6 cells, demonstrating that Gek is broadly required for R cell columnar targeting decisions (supplementary material Fig. S5). Expressing a full-length gek transgene in R cells homozygous mutant for gekomb1080 strongly rescued all targeting defects, as 91% (n=61/67) of axons of this genotype extended correctly, only 3% (n=2/67) mistargeted and 6% (n=4/67) did not extend (Fig. 5E,K,Q,W; supplementary material Fig. S5). Thus, Gek is cell-autonomously required for R1-R6 axons to choose appropriate targets.
We next examined the effect of deleting the regulatory domains of Gek. Surprisingly, expressing either Gek-ΔCC or Gek-ΔPH in R cells homozygous for gekomb1080 was almost as effective as expressing the full-length gek transgene at rescuing the axon targeting phenotypes associated with loss of Gek activity. With Gek-ΔPH, 83% (n=34/41) of axons extended normally, whereas 12% (n=5/41) mistargeted and 5% did not extend (n=2/41; Fig. 5F,L,R,X; supplementary material Fig. S5). Similarly, with Gek-ΔCC, 84% (n=41/49) of axons extended normally, whereas 4% (n=2/49) mistargeted and 12% did not extend (n=6/49) (supplementary material Fig. S5; data not shown). Thus, these gek transgenes retain their normal function in the growth cone. Consistent with the fact that generating single-cell clones using the pan-neural ElavGal4 driver causes much lower levels of transgene expression that that seen using GlassGal4, we did not observe premature stopping of R1-R6 axons above the lamina plexus.
Cdc42 is required for columnar targeting of R1-R6 axons
As Gek was originally identified based on its biochemical association with the small GTPase Cdc42, we tested whether Cdc42 plays a similar role to Gek in R cell axons. Using Cdc424, a loss-of-function allele, we generated R1-R6 cells that were homozygous mutant using the MARCM method, and scored their axonal targeting decisions (Fig. 6; supplementary material Fig. S5). Compared with control R cells, which invariably targeted normally (n=68/68; Fig. 6A-B′,D-E′,G-H′,J-K′; supplementary material Fig. S5), only 26% (n=21/82) of Cdc42 mutant R cells targeted normally, whereas 23% (n=19/82) selected inappropriate target columns and 51% (n=42/82) failed to extend (Fig. 6C,F,I,L; supplementary material Fig. S5). These defects were thus quantitatively and qualitatively indistinguishable from those associated with mutations in gek. Intriguingly, expression of the Gek-ISP transgene fully rescued the gek mutant phenotype (supplementary material Fig. S5). We infer that although physical association between Gek and Cdc42 occurs in a strictly CRIB-dependent fashion in vitro, either other proteins that are present in vivo allow them to interact independently of this domain, or their direct association is not required for their function in R cell growth cones. Moreover, Gek is not the only downstream target of Cdc42, as overexpression of a wild-type gek transgene did not rescue the Cdc42 mutant phenotype (supplementary material Fig. S5). Finally, although the targeting phenotype observed in R cell axons mutant for Cdc42 was specific, these mutant photoreceptors sometimes displayed morphological defects in the retina (data not shown), consistent with Cdc42 playing a broader role in R cell development. These defects were not, however, correlated with the axon targeting phenotype (data not shown).
Cdc42 and Gek regulate a common target in R cell growth cones
Homologs of Cdc42 and Gek can regulate the phosphorylation of the regulatory subunit of myosin light chain (MRLC), affecting actin-myosin contractility in non-neuronal cells (Dong et al., 2002; Wilkinson et al., 2005). Using an antibody that specifically recognizes a phosphorylated form of MRLC in Drosophila (Lee and Treisman, 2004), we tested whether Gek phosphorylates MRLC in R cell growth cones. Phosphorylated MRLC could be detected in growth cones in the lamina, displaying strongly punctate staining above a relatively uniform pattern. Clones homozygous mutant for gekomb1080 displayed strongly reduced punctate staining, while retaining the uniform background (Fig. 7A-A″ ). Consistent with the apparent gain-of-function effect of deleting the PH domain, expression of Gek-ΔPH in gek null cells increased phospho-MRLC staining beyond that in wild-type cells (Fig. 7B-B″ ).
We also observed that clones homozygous for Cdc424 displayed a subtle reduction in phospho-MRLC staining, consistent with the notion that Gek and Cdc42 regulate this common target (Fig. 7C-C″ ). We emphasize, however, that gek mutant clones displayed stronger defects in phospho-MRLC staining than Cdc42 mutant clones, consistent with other upstream mechanisms regulating Gek.
We next reasoned that if Gek or Cdc42 were regulated by the cell surface molecules known to be required for R1-R6 targeting, phosphorylation of MRLC should also be altered in these mutants. However, staining R cell clones homozygous for mutations in the classical cadherin N-cadherin (Ncad), the receptor tyrosine phosphatase Lar, the atypical cadherin flamingo, or both N-cadherin and Lar, did not indicate reductions in phospho-MRLC staining during mid-pupal development (Fig. 7D-G″ ). As a further test of the idea that Gek might function downstream of these molecules, we expressed a Gek-ΔCC transgene in R cells homozygous for mutations in N-cadherin or Lar, and observed no phenotypic rescue (data not shown). Thus, although Cdc42 and Gek regulate MRLC in R cell growth cones, the cell surface proteins known to be important for R cell target selection are not individually necessary for Cdc42 and Gek to do so. As mutations in spaghetti squash, the homolog of MRLC, cause severe defects in early R cell morphogenesis (Lee and Treisman, 2004), precluding analysis of axon targeting, functional tests of this regulation await temporally targeted manipulations of gene activity.
Mutations in gek affect R cell growth cone morphology
Given the phenotypic similarities between gek and Cdc42 mutant clones, as well as the effects on phosphorylation of MRLC, a cytoskeletal regulator, we examined whether Gek regulates growth cone morphology. We examined single growth cones in control and gek mutant R cells at a stage prior to target selection in the lamina, and observed that although gek mutant growth cones appear qualitatively similar to control growth cones, they are quantitatively different (supplementary material Fig. S6). In particular, gek mutant growth cones displayed 20% fewer filopodia and were ∼30% smaller than corresponding control growth cones (supplementary material Fig. S6). These data are consistent with Gek regulating cytoskeletal dynamics in R cell growth cones, affecting their ability to respond to targeting cues.
Gain-of-function mutations in gek alter its subcellular activity
Each R cell body has a unique cross-sectional profile, defined by a developing rhabdome located at a central position within the ommatidium. In photoreceptors homozygous for gekomb1080, these cell type-specific morphologies were invariably normal (n=61/61; Fig. 8A,B,I). Similarly, in gekomb1080 cells expressing a full-length gek transgene, virtually all R cell bodies displayed a normal morphology (99%, n=83/84), with a single cell (n=1/84) displaying an abnormal cross-sectional profile (Fig. 8C,D,I). Thus, Gek activity is not required for photoreceptor cell body morphogenesis, and overexpression of an intact Gek protein does not disrupt this process. However, in R cells that were homozygous for gekomb1080 and expressed either Gek-ΔPH or Gek-ΔCC, we observed two types of defects. First, in both cases, ∼18% of R cells (n=21/88 for Gek-ΔPH and n=16/86 for Gek-ΔCC) displayed abnormal morphologies in their cross-sectional profile, with apical domains extending either too far towards the center of the ommatidium or being located basally (Fig. 8E-I). In addition, ∼6% (n=5/88 for Gek-ΔPH and n=6/86 for Gek-ΔCC) of all R cell bodies had dropped below the plane of the ommatidium (Fig. 8G,I). All remaining R cells in both genotypes were normal. Photoreceptor morphogenesis phenotypes could not be detected in third instar eye discs using GlassGal4 to overexpress Gek-ΔCC, Gek-ΔPH or Gek-ΔCH, although adult animals of these genotypes had rough eyes (data not shown). We infer that these regulatory domains normally prevent Gek from interacting with the molecular machinery that drives R cell morphogenesis in the pupal retina.
Overexpression of full-length Gek induces early termination of R7 axons
To test whether high levels of wild-type Gek expression are sufficient to affect the targeting decisions of other R cell subtypes, we overexpressed full-length Gek in all R cell axons and visualized R7 and R8 cells in adult animals using the opsin transgenes Rh4-lacZ and Rh6-lacZ, respectively. In control experiments in which mCD8GFP was overexpressed using GlassGal4, 1.5% (n=393) of R7 cells targeted to the R8 layer, whereas 5.4% (n=1178) of R8 axons extended aberrantly (Fig. 9A-C). By contrast, 28.3% (n=1278) of R7 cells expressing full-length Gek targeted either to the R8 layer or to an even more superficial layer in the outer medulla, or occasionally to the lamina (Fig. 9D,E). R8 axon targeting was unaffected in this genetic background (4.9%, n=857; Fig. 9F). Thus, increasing Gek activity can cause premature termination of axon extension, leading to targeting errors in R7, but not R8, cells.
Our data demonstrate that Gek and Cdc42 function as regulators of the actin cytoskeleton within R1-R6 cell growth cones. R1-R6 growth cones lacking Gek or Cdc42 activity display two prevalent targeting phenotypes, as they can either fail to extend to their terminal targets or can extend to one or more inappropriate targets within the lamina. Taken together with the biochemical studies of MRCK, we propose that localized activation of Gek by Cdc42 within subdomains of R1-R6 growth cones is a necessary step in making a directed extension towards a specific target. In this view, Gek acts downstream of adhesive cues and/or targeting signals generated on the cell surface, integrating these cues and altering the organization of the actin cytoskeleton so as to promote lateral axon extension. gek mutant growth cones that fail to extend presumably fail to define an extension domain; those that mistarget or that extend to multiple targets must initiate extensions in multiple, spatially inappropriate domains. These latter cases might reflect perduring Gek or Cdc42 protein in our single-cell clones, or might result from the activities of other regulators of protrusion. Our data also demonstrate that the known regulators of R cell axon targeting, including the cadherins Flamingo and N-cadherin as well as the tyrosine phosphatase Lar, are not individually required to regulate Gek or Cdc42. Thus, either these regulators act redundantly to control Gek, or an as yet unknown pathway is required for columnar targeting of R1-R6 growth cones.
Our structure-function analyses of Gek, combined with biochemical studies examining MRCK regulation in vitro, provide insight into the mechanism of Gek regulation. Work in both contexts demonstrates that the kinase domain is the crucial effector of the protein, as deletion or inactivation of this domain removes all detectable activity both in vivo and in vitro (this work) (Luo et al., 1997; Leung et al., 1998). Previous studies of MRCK uncovered multiple negative regulatory mechanisms (Tan et al., 2001a; Tan et al., 2001b; Dong et al., 2002). In particular, the coiled-coil domain drives oligomerization and contains a kinase-inhibitory motif (KIM) that sequesters the kinase domain when the protein is inactive (Tan et al., 2001a; Dong et al., 2002; Garcia et al., 2006). Inhibitory activity has also been associated with the pleckstrin homology domain and the citron homology domain (Chen et al., 1999). Consistent with these observations, when we delete any one of these three domains from Gek and express these constructs in gek mutant R cells, we observe premature stopping of R cell axons above the lamina or defects in cell morphogenesis, depending on the level and timing of transgene expression. However, these mutant forms of Gek do not correspond to simple ‘hyperactivated’ proteins, as adding additional wild-type Gek protein eliminates, rather than enhances, the premature axon termination phenotype. Thus, the presence of wild-type protein restores normal inhibition to these mutant forms of Gek. These results are consistent with biochemical studies of MRCK showing that a conserved N-terminal domain can mediate dimerization, and that kinase activation is critically dependent on transautophosphorylation (Tan et al., 2001a). In this model, activation of a Gek dimer requires that both monomers are independently disinhibited, making them competent to be transphosphorylated by their partner. Thus, in a heterodimer comprising one wild-type subunit and one subunit lacking an autoinhibitory domain (Gek-ΔCC, Gek-ΔPH or Gek-ΔCH), the truncated monomer cannot phosphorylate its inactive wild-type partner (which remains autoinhibited) and does not get phosphorylated by it in turn. As a result, the presence of wild-type protein will suppress an otherwise gain-of-function mutant protein. Conversely, in the absence of wild-type protein, homodimers comprising only truncated subunits will transactivate throughout the cell, making growth cones competent to extend laterally, but also disrupting morphogenesis. Of course, because we cannot directly measure Gek activity within R cell growth cones, we cannot exclude the possibility that the gain-of-function gek mutations that we have generated act differently from the inhibitory activities that have been associated with these domains in MRCK in vitro.
Of the many molecules known to be involved in R cell axon target selection, Gek is the first to be specifically required for R1-R6 axons, but not R7 or R8 axons, to reach their targets. A simple mechanism to achieve such functional specificity would be cell type-specific expression of Gek. However, Gek is strongly expressed in all photoreceptor axons at developmentally appropriate stages. Thus, cell type-specific function must emerge through post-translational regulation of Gek activity. We propose that the cell type-specific requirement for Gek emerges from differences in how upstream signaling pathways control common regulators of the actin cytoskeleton to achieve columnar versus laminar targeting.
MRCK function has been linked to two important regulators of the actin cytoskeleton, LIMK and MRLC, the latter through both direct and indirect pathways Gally et al., 2009; Kumfer et al., 2010; Wilkinson et al., 2005; Dong et al., 2002; Sumi et al., 2001; Tan et al., 2001b). Our data are consistent with the notion that at least one of these, MRLC, is a regulatory target of both Gek and Cdc42 in R cell growth cones, albeit one of as yet untested functional significance. We propose that the cell type-specific requirement for Gek in R1-R6 axons reflects the fact that an as yet unknown upstream signaling pathway requires a specific pattern of modulation of MRLC (achieved via Gek and Cdc42) for R1-R6 axons to be competent to achieve columnar target selection. In this view, the various cell surface interactions that are required for layer-specific targeting of R7 do not signal through Gek, but do use downstream effectors that can be modulated by Gek. When Gek is overexpressed in R7, mistargeting is caused by high levels of Gek bypassing this upstream regulation, causing abnormal activation of relatively ‘generic’ effector proteins in R7 growth cones. Thus, the columnar targeting of R1-R6 axons and the layer-specific targeting of R7 axons reflect the modulation of at least some of the same effector genes, but the upstream signals that direct columnar targeting are uniquely dependent on Gek function. More broadly, given that axon targeting in most systems is generally dependent on multiple, partially redundant cell surface interactions, whereas many of the direct cytoskeletal regulators (such as MRLC) are essential and pleiotropic, our results demonstrate that one way by which cell type-appropriate differences in targeting can be achieved is through the use of specific cytoskeletal modulators like Gek. In this way, subtle differences in the control of the cytoskeleton can cause significant differences in the overall pattern of target selection, making one group of cells competent to target in a columnar fashion. Thus, cell type-specific morphologies can emerge both through differences in cell surface interactions and through differences in cytoplasmic signaling mechanisms.
We thank Heather McFarlane, Victoria Ahn, Richard Reimer, Takashi Suzuki and William Weis for guidance in purifying Gek protein; the Drosophila Genetic Resource Center (Kyoto Stock Center), the Bloomington Drosophila Stock Center and the Drosophila Genomics Resource Center for reagents; and Marion Silies and Jessica Tsai for helpful comments on the manuscript.
This work was supported by the National Institutes of Health [R01 EY015231 to T.R.C.]. Deposited in PMC for release after 12 months.
Competing interests statement
The authors declare no competing financial interests.
Supplementary material available online at http://dev.biologists.org/lookup/suppl/doi:10.1242/dev.069930/-/DC1
- Accepted September 8, 2011.
- © 2011.