A genetic screen for SH2/PTB family genes that synergize with eya during photoreceptor axon targeting

We previously showed that Eya shuttles between the nucleus and cytoplasm and that Abl-mediated tyrosine phosphorylation (pY) can sequester it in the cytoplasm (Xiong et al. 2009). These observations led us to hypothesize that pY–Eya might be recruited to cytoplasmic signaling complexes through interactions with Src homology 2 (SH2) or phosphotyrosine binding (PTB) domain-containing proteins. To test this idea, we leveraged our prior demonstration of genetic synergy between eya and abl during axon guidance (Xiong et al. 2009) to design an RNAi-based genetic interaction screen aimed at identifying SH2/PTB factors relevant to Eya function in differentiating photoreceptors.

Our prior work showed that in viable hypomorphic eya mutants, photoreceptor axons fasciculate aberrantly and frequently overshoot the lamina (Xiong et al. 2009). MARCM clones ruled out the possibility that these phenotypes stemmed from defects in the target layers of the brain, and GMR–Gal4-driven eya knockdown further confirmed its requirement in the photoreceptors (Xiong et al. 2009). Although the GMR > Eya^RNAi mistargeting phenotype was modifiable, as evidenced by its enhancement upon abl knockdown (Xiong et al. 2009), we opted against using it as the primary background for our screen. Briefly, because we predicted positive genetic interactions between eya and relevant SH2/PTB genes, and therefore enhancement of the GMR > Eya^RNAi phenotype, we were concerned that the difficulty in distinguishing a relevant enhancement from a nonspecific additive interaction would lead to a high false positive rate.

As an alternative, we explored the suitability of an eya overexpression phenotype for the screen. Using the membrane marker Ro-lacZ^tau (RTL) to label photoreceptors R2–R5 (Garrity et al. 1999), we quantified the average number of axon bundles that mistarget past the lamina to the medulla in GMR-Gal4 > UAS–Eya^WT animals as 25.7 ± 6.1 (mean ± SD), a significant increase over the driver-alone baseline of 8.1 ± 2.2 (Figure 1, B vs. A and D) (Xiong et al. 2009). Reducing endogenous eya dosage with heterozygosity strongly suppressed the mistargeting (9.2 ± 4.4), confirming that the GMR > Eya^WT phenotype resulted from increased Eya activity rather than from a dominant negative effect of eya overexpression (Figure 1, C and D).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Eya overexpression axon targeting phenotypes provide dose-sensitive, suppressible genetic backgrounds. (A–C) The Ro-lacZ^tau (RTL) marker labels R2–R5 and is visualized by staining for β-galactosidase in third instar larval brains. (A) Wild-type RTL pattern shows most R2–R5 axons terminate correctly at the lamina, whose edges are marked by yellow arrowheads; a few axon bundles overshoot to the medulla, blue arrowhead. (B) eya overexpression causes many photoreceptor axon bundles to overshoot the lamina, marked by yellow arrowheads. (C) eya heterozygosity suppresses eya overexpression defects. (D) Quantification of genetic interaction tests, demonstrating eya overexpression phenotypes are dose sensitive and suppressible. N values are shown in parentheses. (E) Thirty-six SH2/PTB domain-containing genes in Drosophila, grouped into known categories of signaling, that were targeted in our RNAi-based genetic screen.

Although the similarity between eya loss- and gain-of-function phenotypes may seem surprising, there is precedent for biological processes, including photoreceptor axon guidance, being sensitive to precisely balanced gene dosage such that both loss and gain produce similar phenotypes. For example, analogous to our observations with eya, either increased or decreased activity of the p21-activated kinase Pak (Hing et al. 1999) or of the transcription factor hindsight (Oliva and Sierralta 2010) causes photoreceptor axon mistargeting. It is also important to note that our system of counting the number of RTL-labeled axons that overshoot their laminar target, although highly effective as a first-pass quantification of a phenotype or interaction, ignores qualitative differences. For example, eya knockdown tends to produce thicker fascicles and greater disruptions to the lamina plexus than eya overexpression (Figure 1B, Supplemental Material, Figure S1) (Xiong et al. 2009). Future detailed mechanistic studies will be required to understand the requirement for balanced Eya levels in photoreceptor axon guidance, and the eya interactors identified in our screen should provide useful tools in that endeavor.

Given the general caveats associated with overexpression, it was critical to confirm the GMR > Eya^WT mistargeting phenotype could be appropriately modified by a verified eya interacting gene. Building on our prior work (Xiong et al. 2009), we showed that reduced abl dose, either by heterozygosity or by expression of a weak UAS–Abl^RNAi transgene that on its own did not produce targeting defects, dominantly suppressed GMR > Eya^WT (13.4 ± 3.5 and 13.5 ± 5.4, respectively; Figure 1D shows data for Abl^RNAi). Thus reduced abl dominantly enhances eya loss-of-function and dominantly suppresses eya gain-of-function-induced axon mistargeting. We conclude that despite both overexpression and knockdown of eya causing axons to overshoot the lamina, these genetic backgrounds produce genetically distinct and biologically informative phenotypes and interactions.

We also tested sensitivity of the GMR > Eya^WT background to enhancement by coexpressing a second UAS–Eya^WT transgene that had a comparably strong phenotype as the first (24.0 ± 5.4), but found only a modest increase to 30.3 ± 4.1 mistargeted axon bundles per brain. Thus the overexpression phenotype was dose sensitive toward suppressive interactions, but was not easily enhanced (Figure 1D). Because we predicted that reducing the dosage of an SH2/PTB-containing factor relevant to cytoplasmic Eya functions would suppress rather than enhance, the GMR > Eya^WT background seemed ideal for the screen.

We screened a collection of 63 RNAi transgenes that targeted 36 SH2/PTB family genes for their ability to suppress the GMR > Eya^WT axon targeting phenotype (Figure 1E). This panel included two independent RNAi transgenes per gene whenever possible to minimize the likelihood of false positives from off-target effects and false negatives from insufficient knockdown. Our use of GMR–Gal4 restricted knockdown to retinal cells posterior to the morphogenetic furrow, thereby bypassing any earlier requirements for the SH2/PTB genes to be tested. The screen was performed blind as to SH2/PTB gene identity, and ∼10–20 third larval instar eye–brain complexes were dissected, stained, and scored for each interaction test.

To prioritize further study, suppressors were binned into strong, moderate, or mild interest groups based on P-value. A total of 11 strong (P < 0.001), 6 moderate (P < 0.01), and 10 mild (P < 0.05) hits were recovered, identifying 21 potential eya interacting genes (Table 1 shows genes with one strong or moderate hit; full screen results are in Table S1). The rather high hit rate from the primary screen may reflect extensive connectivity between phosphotyrosine signaling modules (Koytiger et al. 2013). Inclusion of dab, a positive factor in the Abl signaling network (Song et al. 2010), in this initial list provided proof of principle for the screen by highlighting the importance of eya–abl interactions to axon guidance and thereby further validating the biological relevance of the eya overexpression phenotype. To narrow the scope of the follow-up experiments to a technically manageable level, we focused only on the strong and moderate suppressors, with one exception, Stat92E.

Genetic triage highlights potential links between Eya and both the Jak/Stat signaling pathway and nonreceptor tyrosine kinases

Three secondary genetic tests were used to verify these hits as bona fide genetic interactors and to determine which genes may be relevant to cytoplasmic Eya function. First, we asked whether lines that suppressed GMR > Eya^WT conversely enhanced eya loss-of-function axon targeting phenotypes (Table 1). All but three genes (CG15529, plx, and vap) enhanced GMR > Eya^RNAi, while shark yielded mixed results, as one shark RNAi line enhanced and the other suppressed. Next, we used membrane-tethered myristoylated Eya (Myr–Eya) and nuclearly localized Eya (NLS–Eya) (Xiong et al. 2009) to distinguish interactions specific to Eya’s cytoplasmic and nuclear functions, respectively. GMR–Gal4-driven expression of both Myr–Eya and NLS–Eya produce axon guidance defects that can be suppressed by eya heterozygosity (Figure 1D), indicating gain-of-function effects. However, we hypothesized that Myr–Eya and NLS–Eya perturb axon targeting via distinct mechanisms given their restricted subcellular localizations. Consistent with this idea, abl knockdown suppressed the axon mistargeting phenotypes associated with Myr–Eya but not NLS–Eya (Figure 1D). Of the candidate lines tested, all except csw and X11Lβ significantly suppressed GMR > Myr–Eya mistargeting defects (Table 1). When tested against GMR > NLS–Eya, only Src64B and csw suppressed, while the remaining 10 lines enhanced. Because relevant interactions with eya gain-of-function phenotypes should manifest as suppression, we suspected the enhancement of GMR > NLS–Eya reflected additive effects rather than real interactions. Consistent with this prediction, GMR–Gal4 knockdown of these SH2/PTB genes produced significant targeting defects (Figure 2A). In sum, the results of our secondary screens whittled the number of candidates to 10 genes: Csk, hop, Lnk, shark, Socs36E, Socs44A, Src42A, Src64B, Stat92E, and X11L. Among these, Src64B was unique in its potential to influence both cytoplasmic and nuclear Eya functions.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Eya-interacting SH2/PTB genes are required for photoreceptor axon guidance. (A) Quantification of axon mistargeting phenotypes resulting from RNAi knockdown of 10 SH2/PTB genes from Table 1. The RNAi lines indicated were crossed to GMR;RTL. Superscripted numbers indicate the VDRC stock used. Numbers shown are averages of overshooting axons per brain with SD. * P-value <0.05, ** P-value < 0.01, *** P-value < 0.001. N values are shown on the right. (B–F) Representative brains of the indicated genotype stained with β-galactosidase.

Two groups of genes were notably enriched in this short list: the Jak/Stat pathway (hop, Stat92E, Socs36E, and Socs44A) and nonreceptor tyrosine kinases (Csk, hop, shark, Src42A, and Src64B). In the follow-up studies described below, we focused on uncovering the mechanistic links that may underlie the Jak/Stat signature. Two reasons drove this decision. First, because the Jak/Stat pathway had not been previously implicated in axon guidance in any system, discovery of a novel context for Jak/Stat signaling in neuronal development had the potential for the broadest impact. Second, with the exception of Stat92E, the interactions that produced our 10-gene short list were based on a single RNAi transgene, either because only one line was available (Csk, Src42A, Src64B, and X11L) or because only one of the two lines tested showed interaction in the primary (hop, Lnk, Socs36E, and Socs44A) or secondary (shark) screens (Table 1, Table S1). Independent identification of four different Jak/Stat pathway members provided collective validation, and so mitigated the risk of pursuing spurious off-target effects of single hits. In the future, comparison of target gene messenger RNA levels between pairs of lines and addition of Dicer2 to the line that did not interact could be performed to determine whether such discrepancies reflect false positives due to off-target effects or false negatives due to either poor RNAi efficiency or additive interactions. We consider the latter to be the most likely explanation, given that our screen required suppression of GMR > Eya^WT by RNAi knockdowns that on their own often produced strong mistargeting defects (Figure 2A). Thus a “hit” in the primary screen often reflects not only suppression of eya overexpression phenotypes by SH2/PTB knockdown, but also mutual suppression of SH2/PTB knockdown phenotypes by increased eya dose. In cases where the additive effects were sufficiently strong as to mask such mutual suppression, inconsistent results between independent lines would be expected.

The second category of interactors, the nonreceptor tyrosine kinases, underscores the importance of phosphotyrosine-based signaling for photoreceptor axon targeting. Previous work hinted at this, as a number of tyrosine phosphatases were found to be critical for regulating axon targeting (Garrity et al. 1999; Schindelholz et al. 2001; Jeon et al. 2008). Our results identify a set of candidate tyrosine kinases that could provide the opposing catalytic activity. The remaining two genes on the short list, Lnk and X11L, encode adapter proteins. X11L has been implicated in amyloid precursor protein (APP) regulation in both Drosophila and mammalian cells (Hase et al. 2002; Gross et al. 2008), with no reported connections to either Jak/Stat or tyrosine kinase signaling. Lnk negatively feeds back on Jak/Stat signaling in mammalian cells, although in Drosophila it has been studied as a positive regulator of insulin/insulin-like growth factor signaling (Werz et al. 2009; Oh et al. 2010; Slack et al. 2010; Almudi et al. 2013). The Jak/Stat ligands unpaired (Upd) 2 and 3 influence insulin signaling in Drosophila (Rajan and Perrimon 2012; Woodcock et al. 2015), thus potentially implicating Lnk in Jak/Stat signaling in flies.

Jak/Stat signaling is necessary for photoreceptor axon guidance

Jak/Stat signaling in Drosophila is transduced via three ligands, Upd1–3, one receptor, Dome, one Janus kinase (JAK), Hopscotch (Hop), and one signal transducer and activator of transcription (STAT), Stat92E (Stat). Signaling can be attenuated by a variety of negative regulators, including the suppressor of cytokine signaling (SOCS) proteins encoded by Socs36E and Socs44A (Rawlings et al. 2004a,b; Arbouzova and Zeidler 2006). Jak/Stat signaling directs a diverse spectrum of developmental events including cell migration, proliferation, patterning, and morphogenesis (reviewed in Arbouzova and Zeidler 2006 and Li 2008). During eye development, the pathway promotes growth of the early eye field, contributes to initiation and movement of the morphogenetic furrow and to regional eye specification, transduces information from the dorsal–ventral axis to direct ommatidial rotation, and regulates specification and maintenance of the optic lobe neuroepithelium (Luo et al. 1999; Zeidler et al. 1999; Chao et al. 2004; Tsai and Sun 2004; Ekas et al. 2006; Gutierrez-Aviño et al. 2009; Ngo et al. 2010; Wang et al. 2011).

Identification of hop, Stat92E, Socs36E, and Socs44A in our genetic screen (Table 1) and finding that GMR-driven knockdown of these genes produces axon mistargeting phenotypes (Figure 2, A–E) suggested that the Jak/Stat pathway is also active in postmitotic photoreceptors. In support of this, expression of the 10xStat–eGFP pathway activity reporter suggests that early expression of the ligand, upd, which is absent by the third instar, enables sustained Jak/Stat signaling in the photoreceptors at the time they are sending out their axons (Bach et al. 2007). As predicted, GMR–Gal4-driven knockdown of the receptor dome, but not of the upd ligands, produced significant mistargeting defects (Figure 3, A–C).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Jak/Stat signaling is required cell autonomously in photoreceptors for axon targeting. (A) Quantification of axon mistargeting phenotypes, resulting from RNAi knockdown of upd and dome. The RNAi lines indicated were crossed to GMR;RTL. Superscripted numbers indicate the Bloomington stock used. Numbers shown are averages of overshooting axons per brain with SD. * P-value <0.05, ** P-value < 0.01, *** P-value < 0.001. N values are shown to the right. (B and C) Representative brains of the indicated genotype stained with β-galactosidase. (B) upd knockdown does not perturb axon targeting. (C) Knockdown of dome leads to significant overshooting of the lamina. (D) GFP⁺ (green) MARCM clone showing that the axons of Stat92E null photoreceptors overshoot the lamina. Chaoptin staining (red) marks all photoreceptors in the eye disc (top right) and highlights their axonal projections through the optic stalk and into the brain. (E) Representative brain stained with β-galactosidase, showing that Socs36E knockdown suppresses the axon mistargeting phenotypes of GMR > hop^RNAi (compare to Figure 2B) (F) Quantification and statistics of axon overshooting phenotypes in double knockdown interaction assays show that Socs genes function as Jak/Stat pathway antagonists. GMR;RTL (blue), GMR > Socs36E^RNAi;RTL (red), and GMR > Socs44A^RNAi;RTL (green) were crossed to the hop^RNAi line indicated on the x-axis. N values per experiment from left to right are 10, 7, 10, 13, 13, and 14. P-values were calculated by performing Student’s t-test between the experimental cross and driver-alone control (i.e., red/green vs. blue). * P-value <0.05, ** P-value <0.01, *** P-value <0.001. Superscripted numbers refer to VDRC RNAi line.

To confirm further the requirement for Jak/Stat signaling in photoreceptor axon guidance, we generated MARCM clones of a Stat92E null allele, stat^85c9. Axons originating in photoreceptors that lacked Stat92E overshot the lamina into wild-type brain tissue (Figure 3D), while clones within the brain did not cause mistargeting of wild-type photoreceptors (data not shown). This demonstrates that correct axon targeting requires Stat92E in the photoreceptors. The result also eliminates the possibility that RNAi-induced phenotypes resulted from either leaky GMR–Gal4 expression in the brain (Li et al. 2012) or off-target effects. Together, our results indicate a cell autonomous role for Jak/Stat signaling in postmitotic photoreceptors that directs axon projections to the correct layers of the brain.

One apparent inconsistency between this conclusion and our genetic screen results is that SOCS factors typically antagonize Jak/Stat signaling (Chen et al. 2000; Callus and Mathey-Prevot 2002; Rawlings et al. 2004b) yet knockdown of Socs, hop, or Stat92E all suppressed GMR > Eya^WT. Since axon targeting is a novel context for Jak/Stat signaling, one possibility is that the SOCS factors do not antagonize the pathway in postmitotic photoreceptors as they do in other situations (Callus and Mathey-Prevot 2002; Rawlings et al. 2004b). Arguing against this, knockdown of either Socs36E or Socs44A suppressed the GMR > hop^RNAi mistargeting phenotypes (Figure 3, E and F). In addition to their roles as negative regulators of Jak/Stat signaling, SOCS factors also influence a variety of growth factor signaling cascades (De Sepulveda et al. 1999; Chen et al. 2000; Bayle et al. 2004; Rawlings et al. 2004b; Kazi et al. 2012; Trengove and Ward 2013). Thus the eya–Socs synergy observed in our screen could reflect interactions relevant to those other pathways. Mechanistically, although SOCS proteins belong to the family of Cullin–Ring E3 ubiquitin ligases (reviewed in Piessevaux et al. 2008), a model in which they target Eya for degradation seemed unlikely because the predicted increase in Eya levels upon Socs knockdown should enhance, rather than suppress, GMR > Eya^WT phenotypes. Consistent with such reasoning, Socs knockdown did not alter Eya protein levels in the GMR > Eya^WT background (data not shown).

Given the sensitivity of photoreceptor axon targeting to both loss and overexpression of eya (Figure 1, Table 1, Figure S1), we asked whether the same might hold true for hop, Stat92E, Socs36E, and Socs44A. GMR–Gal4-driven overexpression of hop was lethal prior to the third instar, and Stat92E overexpression did not produce targeting defects (data not shown and Figure 4, A and C). Stat92E overexpression also did not modify GMR > Eya^RNAi (Figure 4, B and F), suggesting endogenous signaling is limited and insufficient to activate the extra Stat protein. Consistent with this interpretation, previous reports have shown that although UAS–Stat can rescue Stat92E null phenotypes, its overexpression in an otherwise wild-type background does not perturb eye development (Ekas et al. 2006).

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

Eya and SOCS synergize in axon targeting. (A) Quantification of axon phenotypes resulting from overexpression of Stat92E, Socs36E, and Socs44A. (B) Quantification of axon phenotypes from genotypes in (A) in a GMR > Eya^RNAi background. (C–H) Representative images of the indicated genotypes stained with anti-β-galactosidase. P-values were calculated by performing Student’s t-test between the experimental cross and the control of that group. In B, the first is relative to GMR > Eya^RNAi and the second is relative to GMR > SOCS^WT, revealing mutual suppression. * P-value <0.05, ** P-value <0.01, *** P-value <0.001.

In contrast, GMR–Gal4-driven overexpression of either Socs36E or Socs44A significantly perturbed axon targeting (Figure 4, A, D, and E). We took advantage of this result to validate independently the Socs–eya synergy predicted by our genetic screen interactions. Just as Socs knockdown suppressed eya overexpression phenotypes (Table 1), Socs overexpression suppressed eya knockdown phenotypes (Figure 4, B, G, and H). Thus cooperative eya–Socs interactions appear integral to photoreceptor axon targeting.

Eya co-immunoprecipitates and colocalizes with Socs44A via an SH2–pY-mediated interaction

Given that the impetus for our genetic screen was the possibility that SH2/PTB domains interact with pY–Eya, we next asked whether Eya could co-immunoprecipitate (CoIP) Jak/Stat proteins from lysates of Drosophila S2 cultured cells transiently transfected with epitope-tagged expression constructs. CoIPs were performed both with and without cotransfection of Abl. We found that Flag-tagged Eya can CoIP HA-tagged Hop, Socs36E, and Socs44A, but not Stat, and that Abl improved Eya’s CoIP of Socs36E and Socs44A, but not that of Hop or Stat (Figure 5, A–D). Abl stimulates Eya tyrosine phosphorylation and cytoplasmic accumulation (Xiong et al. 2009 and Figure 6, A vs. B, quantified in N) and thus should enhance SH2/PTB–pY–Eya interactions. Abl is expressed endogenously in S2 cells (Cherbas et al. 2011), and some tyrosine phosphorylation of Eya can be detected in cells transfected with Eya alone (Tootle et al. 2003), perhaps explaining the low levels of Socs36E and Socs44A CoIP detected when Abl was not supplied by cotransfection (Figure 5, C and D). The Abl independence of the Eya–Hop CoIP (Figure 5A) suggests that either Hop phosphorylates Eya at different tyrosine residues, or that pY–SH2 interactions do not drive the CoIP.

• Download figure
• Open in new tab
• Download powerpoint

Figure 5

pY–SH2 dependence of Eya CoIP with Hop, Socs36E, and Socs44A. (A–F) Western blots showing CoIP results from transfected Drosophila S2 cells. Top panels, blotted with anti-HA, show input levels for the HA-tagged Jak/Stat pathway factor. Middle panels, blotted with anti-Flag, show IP of Flag–Eya. Bottom panels show CoIP of the HA-tagged Jak/Stat pathway protein. Key below indicates factors transfected in each experiment. (A–D) Eya CoIPs full-length Hop, Socs36E, and Socs44A but not Stat. Broad smeary band in lane 4, middle panels, reflects the extensive tyrosine phosphorylation of Eya that occurs in the presence of Abl (Xiong et al. 2009). (A) Flag–Eya CoIPs HA–Hop in the presence or absence of Abl. (B) Flag–Eya does not CoIP HA–Stat. (C and D) Flag–Eya CoIP of Socs36E (C) and Socs44A (D) is increased in the presence of Abl. (E and F) Eya–Socs44A but not Eya–Socs36E requires the SH2 domain of that SOCS factor. (E) Flag–Eya CoIPs all HA–Socs36E mutant constructs. (F) Flag–Eya CoIPs all HA–Socs44A deletion constructs except that lacking the SH2 domain. Black triangles mark IgG bands. Full-length Socs44A (lanes 2 and 4) runs below IgG heavy chain, while Socs44A^ΔNT runs coincident with IgG light chain (lane 3).

• Download figure
• Open in new tab
• Download powerpoint

Figure 6

Eya localizes to cytoplasmic punctate structures when cotransfected with Abl and Socs44A. (A–M) Individual transfected S2 cells showing representative subcellular distribution of Flag–Eya and HA–Socs44A. Nuclei are marked with DAPI. (A) Flag–Eya is predominantly nuclear. (B) Cotransfection of Abl causes cytoplasmic redistribution of Flag–Eya. (C) Full-length HA–Socs44A appears cytoplasmic and membrane associated. (D) Myr–Flag–Eya is tightly membrane associated, with some punctate structures apparent. (E) Cotransfection of Abl does not alter Myr–Flag–Eya distribution. (F and G) Two representative cells showing that cotransfection of Socs44A does not induce a major redistribution of Eya (F), although in a small fraction of cells, bright membrane proximal punctate structures are seen (G). Socs44A appears more tightly membrane associated in cells in which Eya puncta form (G′), with reduced general cytoplasmic accumulation (compare to F′). (H and I) Two representative cells showing Eya cytoplasmic (H) and punctate (I) localization increases in response to cotransfection of Abl and Socs44A. In cells with Eya puncta, Socs44A tends to be strongly membrane associated (I′). (J and K) Redistribution of Eya to membrane-associated puncta requires the SH2 domain of Socs44A. Socs44A^ΔSH2 localizes diffusely throughout the cytoplasm and is not strongly membrane associated (Figure S3L), and localization does not change with cotransfection of Eya (J) or Eya and Abl (K). (L) Socs44A only slightly increases the punctate distribution of Myr–Eya. (M) Cotransfection of Abl and Socs44A results in punctate Myr–Eya distribution in 75% of cells. (N) Quantification of nuclear vs. cytoplasmic distribution of Flag–Eya in experiments A, B, and F–I. N values are shown across the top of the graph. (O) Quantification of cytoplasmic Eya distribution into punctate structures in experiments A, B, and D–M. Punctate classification is a subset of membrane associated (i.e., some cells categorized as having punctate staining also have membrane-associated staining). N values are shown across the top of the graph.

Because Abl increased Eya’s CoIP with Socs36E and Socs44A, we focused on them as the top candidates for forming SH2–pY–Eya complexes. To test this, we generated three deletion mutants that removed either the N terminus (ΔNT), the SH2 domain (ΔSH2), or the Socs Box (ΔSB) and a point mutant (SH2*, Socs36E^R500K, or Socs44A^R214K) that compromises SH2 function in homologous mammalian SOCS proteins (Nicholson et al. 1999). Consistent with expectations, the Socs44A SH2 domain was required for CoIP, while the SH2* point mutant weakened the interaction (Figure 5F, Figure S2). Similar experiments with Socs36E did not identify any single domain as being required for CoIP (Figure 5E), suggesting that the Eya–Socs36E interaction is more complicated, either requiring multiple Socs36E domains to assemble the complex and/or additional proteins to bridge the interactions. Socs44A and Socs36E are both endogenously expressed in S2 cells (Callus and Mathey-Prevot 2002; Zhu et al. 2013) and can CoIP each other (Figure S3). Thus, Socs44A is an appealing candidate to bridge the Eya–Socs36E interaction, as its association with both proteins would confer the sensitivity to Abl seen in the Eya–Socs36E CoIPs (Figure 5C).

The pY–SH2-mediated CoIP of Eya and Socs44A predicted cytoplasmic colocalization. Our previous work has shown that Eya is predominantly nuclear but can be relocalized to the cytoplasm upon phosphorylation by Abl (Xiong et al. 2009 and Figure 6, A and B). Socs44A accumulated in the cytoplasm, with strong association to the plasma membrane (Figure 6, C and F–I; localization of Hop, Stat, and Socs36E is shown in Figure S4). In Eya–Socs44A cotransfected cells, we observed a subtle increase in the proportion of cells with cytoplasmic Eya, while Socs44A localization was not changed (Figure 6, F, G, and N). Occasionally, membrane-associated Eya puncta were observed and these tended to colocalize with Socs44A⁺ spots (Figure 6G). The frequency of cells displaying these punctate structures dramatically increased when Abl was cotransfected (Figure 6, H, I, and O). Confirming the SH2 requirement for Eya recruitment, we did not observe Eya-containing puncta when Socs44A^ΔSH2 was expressed (Figure 6, J, K, and O). To separate Abl’s role in phosphorylating Eya from its role in recruiting Eya to the cytoplasm, we examined the response of Myr–Eya, which is constitutively targeted to the cytoplasmic membrane (Figure 6, D and E). Only when Abl was cotransfected did Socs44A promote Eya punctate localization (Figure 6, L, M, and O). We speculate that these puncta represent Eya–Socs44A complexes and that they are mediated by pY–SH2 interactions.

Concluding remarks

Our genetic screen was initially motivated by the idea that the cytoplasm was a likely site for Eya’s phosphotyrosine phosphatase activity (Xiong et al. 2009), a model which was subsequently validated in mammals (El-Hashash et al. 2012). Recent work has shown that this catalytic function of Eya is dispensable for Drosophila development (Jin et al. 2013). What then might be the role of cytoplasmic Eya? One possibility is that Eya might act as a scaffolding factor that nucleates the formation of specific protein complexes in response to different upstream signals. Because our biochemical experiments relied on interactions between overexpressed proteins in cultured cells, it will be important to confirm the results using endogenous proteins from Drosophila tissues. Another possibility is that Eya might modulate the composition or function of such complexes via its phosphothreonine phosphatase activity. Eya’s threonine phosphatase activity has been implicated in innate immunity in both mammals and Drosophila, and although substrates have not been identified, cytoplasmic activity was implied by the identified interactions (Okabe et al. 2009; Liu et al. 2012). Intriguingly, the Jak/Stat pathway is relevant to innate immunity in both Drosophila and mammals (reviewed in Agaisse and Perrimon 2004 and O’Shea and Plenge 2012). In Drosophila, Jak/Stat signaling regulates hemocyte (blood cell) development and release of Upd cytokines in response to infection (Agaisse et al. 2003; Copf et al. 2011; Minakhina et al. 2011; Woodcock et al. 2015). The tyrosine kinases from our short list of candidates also have potential links to immunity. shark and Src42A are important for Draper-mediated glial phagocytosis of axonal debris (Ziegenfuss et al. 2008) which, if left uncleared, stimulates an immune response, and Src64B overexpression has been shown to be sufficient for inducing an immune response in larva (Williams 2009). Even if Eya’s threonine phosphatase activity is not involved, extending exploration of the interactions we have uncovered in the context of axon guidance to other biological contexts in which the candidate interacting partner is known to function may provide new insight into the cytoplasmic signaling networks in which Eya participates during development and disease.