CoinFLP: a system for efficient mosaic screening and for visualizing clonal boundaries in Drosophila

Many of the genes that regulate development in a variety of organisms have been discovered in genetic screens that utilize the Drosophila eye (St Johnston, 2002). An important advance in the design of these screens was the development of methods that allow the generation of eyes bearing homozygous mutant clones (Golic, 1991; Newsome et al., 2000; Xu and Rubin, 1993). This enabled investigators to restrict the deleterious effects of the mutation to a tissue that was not essential for viability or fertility. As a result, mosaic screens in the Drosophila eye have discovered many genes that function in regulating cell proliferation, cell fate specification and photoreceptor differentiation (e.g. Benlali et al., 2000; Tapon et al., 2001; Treisman, 2001).

We now know that there are mutations in several types of genes that function in cell-cell interactions whose effects could only have been discovered in a tissue consisting of two different types of cells (wild type and mutant). These include important subgroups of genes that regulate cell proliferation and the affinity of cell-cell interactions. These mutant phenotypes become evident by changes in the relative representation of the two populations (Tseng et al., 2007), by overgrowth of one population induced by its interaction with the other (Lee et al., 2002), or by abnormalities in the boundaries that separate the two populations (e.g. smooth rather than irregular) (Yang et al., 1999). Importantly, these abnormalities would not have been evident if the entire eye was composed of mutant tissue.

The Gal4/UAS binary expression system in Drosophila (Brand and Perrimon, 1993; Fischer et al., 1988) has enabled researchers to generate mutant cells in a spatially and temporally restricted manner. In addition, it allows the systematic screening in vivo of thousands of Gal4-dependent UAS lines for mutant phenotypes, either by RNA-mediated interference (RNAi) or gene overexpression. Although RNAi and overexpression screens have been extremely effective for identifying genes in a number of complex biological processes in Drosophila (e.g. Colombani et al., 2012; Neely et al., 2010; Pospisilik et al., 2010), it is currently difficult to use this approach in the context of a mosaic screen.

Here we describe a new method, CoinFLP, that allows us to conduct mosaic screens in Drosophila tissues using RNAi or gene overexpression. This system exploits a stochastic recombination choice by the FLP enzyme to generate reliable ratios of mutant and wild-type cells in a developmentally controlled manner. We have validated this approach in the adult eye with a screen of genes that cannot be studied using the commonly used FLP/FRT stocks. Additionally, to facilitate the study of interactions between clones of different mutant genotypes, we have generated versions of CoinFLP that allow the simultaneous visualization of Gal4-expressing and LexGAD-expressing clones in the imaginal disc as well as the boundary between them.

In order to generate Drosophila tissues that are mosaic for expression of a gene of interest, we designed a system that enables the expression of Gal4 from patches of cells in a tissue and excludes its expression in the rest of the animal. We chose a FLP/FRT-based recombination approach to take advantage of the availability of a number of FLP-expressing stocks generated by different groups. The FLP recombinase binds to 11 bases of the inverted repeat sequence in the FRT element but only to the two outermost bases of the 8 bp spacer (Qian and Cox, 1995). The spacer, however, is involved in DNA-DNA pairing during the recombination event. Therefore, the central six nucleotides in the spacer can be mutated to alter the specificity of recombination. Thus, canonical FRT sites (shown as FRT) and FRT3 sites (Fig. 1A) are each capable of undergoing a recombination event with a site of the same kind but not with the other type of site (Schlake and Bode, 1994; Senecoff and Cox, 1986). This type of mutually exclusive recombination between different recombinase target variants has been exploited in several genetic systems, such as iMos (Claveria et al., 2013), Brainbow (Cai et al., 2013; Hadjieconomou et al., 2011; Hampel et al., 2011; Livet et al., 2007), recombinase-mediated cassette exchange (RMCE) (Turan et al., 2013) and the flip-excision (FlEx) transgene inversion strategy (Schnutgen et al., 2005).

We designed and constructed a system that we have named CoinFLP. The first version that we constructed was CoinFLP-Gal4 (Fig. 1B), in which recombination between canonical FRT sites results in the excision of a transcriptional STOP cassette, such that the Gal4 gene is expressed from the Actin 5C (Act) promoter in that cell and all of its descendants. By contrast, recombination between the two FRT3 sites results in the excision of one of the canonical FRT sites, thus preventing any subsequent FLP/FRT-mediated recombination and maintaining the STOP cassette upstream of Gal4 in that cell and all its descendants. The two types of recombination event are mutually exclusive and the choice between them is stochastic, similar to flipping a coin (hence ‘CoinFLP’). The CoinFLP-Gal4 transgene differs from the widely used FLP-out transgenes, such as Act>CD2>Gal4 and Act>y+>Gal4 (Ito et al., 1997; Pignoni and Zipursky, 1997; Struhl and Basler, 1993), in which recombination between FRT sites always results in Gal4 expression.

To generate mosaicism via CoinFLP-Gal4, we expressed FLP recombinase from a separate transgene. One FLP transgene that we used was under the control of the eyeless promoter (ey-FLP) (Newsome et al., 2000). This restricts recombination events to the developing eye-antennal imaginal disc, which gives rise to the adult eye and antenna (Fig. 1C). We empirically identified an ey-FLP transgene whose recombination with CoinFLP-Gal4 was most restricted to the eye-antennal disc (Fig. 1D; supplementary material Table S1, Fig. S1A). However, we observe rare recombination events in fat body and gut cells (supplementary material Fig. S1B,C). We also tested another FLP transgene, Ubx-FLP, which is less restrictive than ey-FLP and induced recombination in wing, leg and eye imaginal discs (supplementary material Fig. S1D-G). In animals bearing ey-FLP and CoinFLP-Gal4, patches of Gal4-expressing cells can be identified with different UAS-based markers. UAS-GFP fluorescently labels Gal4-expressing clones in the eye-antennal disc (Fig. 1D) and UAS-white^RNAi marks clones as white in the adult eye (Fig. 1E).

To facilitate the generation of marked patches of Gal4-expressing cells in a variety of tissues that are visible externally, we used Ubx-FLP and CoinFLP-Gal4 in combination with UAS-regulated RNAi transgenes that affect pigmentation of the cuticle or the morphology of macrochaetae or wing hairs. UAS-ebony^RNAi marks Gal4-expressing clones with dark pigment in the adult dorsal thorax and head (Fig. 1F), UAS-forked^RNAi causes mutant patches to have bent bristles (Fig. 1G), and UAS-mwh^RNAi marks clones with multiple hairs in the wing (Fig. 1H). By comparison, the Act>y+>Gal4 transgene causes most, if not all, cells expressing FLP to express Gal4 in a non-mosaic uniform pattern (Fig. 1I-M). These results demonstrate that CoinFLP-Gal4 is a versatile tool that can be used to generate and visualize Gal4-expressing clones in a variety of tissues when paired with the desired UAS-regulated marker and tissue-specific FLP.

One of the benefits of traditional mitotic recombination using ey-FLP (Newsome et al., 2000) is that a predictable proportion of mutant cells is generated, which has allowed the identification of mutations that cause a significant deviation from the expected ratio (Tapon et al., 2001). When compared with clones generated by mitotic recombination using ey-FLP (supplementary material Fig. S1H,I), clones generated using CoinFLP-Gal4 are fewer and larger (Fig. 1D and see below). This might be because recombination between FRT sites in cis occurs at an earlier stage of development than mitotic recombination, when there are fewer cells in the eye primordium. Since we do not see many small clones expressing Gal4, this suggests that nearly all eye-antennal disc cells have undergone terminal recombination during embryonic development.

To examine clonal phenotypes using CoinFLP-Gal4 and ey-FLP, crosses were set up as shown in Fig. 2A. To determine if the proportion of Gal4 patches produced is consistent from disc to disc, we first measured the relative area of Gal4-expressing cells in the eye-antennal disc in the absence of a UAS-RNAi transgene. Unlike the Act>y+>Gal4 transgene, which undergoes recombination to express Gal4 in nearly every cell (99%) (Fig. 1I, Fig. 2B), CoinFLP-Gal4 expresses Gal4 in approximately half of the disc (56%) (Fig. 1D, Fig. 2B,C). A variety of loss-of-function mutations have been described previously that elicit different phenotypes in clones generated by FLP/FRT-mediated mitotic recombination when compared with those observed in homozygotes. We tested whether some of these phenotypes could be observed in the adult eye using CoinFLP-Gal4 with ey-FLP and UAS-RNAi lines. We included UAS-dcr2 to increase the efficacy of RNAi (Dietzl et al., 2007). Knockdown of expanded (ex) (Boedigheimer and Laughon, 1993), a negative regulator of growth, in clones clearly demonstrates the growth advantage exhibited by mutant cells over wild-type cells (Fig. 2B,F,G), whereas uniform ex knockdown produces an eye similar in size to that of the wild type (Fig. 2H), although the head capsule is larger (not shown).

Clones of cells heterozygous for mutations in a number of different genes encoding ribosomal proteins (Minute) are eliminated in mosaic imaginal discs by a mechanism known as cell competition (Morata and Ripoll, 1975). By contrast, animals that are entirely Minute develop to the adult stage. Similarly, using CoinFLP-Gal4, patches of cells expressing an RNAi that targets RpL24, a likely Minute gene (Marygold et al., 2007), are underrepresented in the eye-antennal disc (Fig. 2B,I) and are eliminated from the adult eye (Fig. 2J), except sometimes at the posterior margin where cell proliferation ceases at an earlier stage of development. By contrast, knockdown of RpL24 in all the cells of the eye has no apparent effect on cell viability (Fig. 2K), demonstrating that these cells are eliminated only when neighboring wild-type cells are present. Another mutation that causes cells to be eliminated by wild-type cells affects the gene scribble (scrib) (Bilder and Perrimon, 2000). As expected, knockdown of scrib in clones results in the elimination of mutant cells (supplementary material Fig. S2C), whereas uniform knockdown results in lethality during development (supplementary material Fig. S2D), which is likely to be due to continued cell proliferation of the eye-antennal discs.

Another advantage of the CoinFLP-Gal4 system over FLP/FRT-induced mitotic recombination is that it allows the examination of phenotypes elicited by the overexpression of genes in clones. Overexpression of reaper (rpr) (White et al., 1994), a pro-apoptotic gene, in clones demonstrates that mutant cells are eliminated while the surviving cells are able to generate an adult eye of normal shape and size (supplementary material Fig. S2E). By contrast, rpr overexpression in the entire eye-antennal disc ablates the head (supplementary material Fig. S2F). Overexpression of Myc (diminutive – FlyBase), a transcription factor, in clones dramatically increases the relative representation of Gal4-positive tissue (supplementary material Fig. S2G), consistent with the ability of Myc to promote growth in a cell-autonomous fashion (Johnston et al., 1999) and to inhibit the survival of neighboring cells that contain lower levels of Myc (de la Cova et al., 2004; Moreno and Basler, 2004). Myc overexpression uniformly in the eye results in a slight overgrowth phenotype (supplementary material Fig. S2H).

Importantly, we find that the CoinFLP-Gal4 system can reveal mutant phenotypes in somatic mosaics that can differ from those observed under conditions of uniform expression. Also, in some cases these phenotypes cannot be visualized under conditions of uniform expression, especially when it results in lethality at an earlier stage of development. Additionally, CoinFLP can be used to examine mosaic phenotypes both in the context of knockdown and of overexpression and can therefore be used for screens that employ collections of RNAi lines (Dietzl et al., 2007; Ni et al., 2011) or collections of cDNAs expressed under UAS control (Bischof et al., 2013).

Screens using FLP/FRT-induced mitotic recombination are unable to recover mutations in genes that are located proximal to the FRT sites, as well as genes on the fourth and Y chromosomes (supplementary material Fig. S3A). In order to assess the potential of CoinFLP for identifying new genetic pathways involved in cell-cell interactions, we built a CoinFLP-Gal4 tester stock that contains ey-FLP and UAS-white^RNAi, and set up crosses with 234 UAS-RNAi stocks corresponding to 175 genes (Fig. 3A; supplementary material Fig. S3A, Table S2). In parallel, we compared the effects of uniform knockdown in a different tissue by expressing each UAS-RNAi in the developing wing pouch with nubbin-Gal4 (nb-Gal4) (supplementary material Table S2).

We observed eye abnormalities in 57 of 234 stocks screened (24%) (Fig. 3B; supplementary material Table S2). Among the genes that scored positive in the screen was cubitus interruptus (ci) (Orenic et al., 1990), which caused overgrowth especially in the anterior part of the eye (Fig. 3E). Signaling mediated by Smoothened (Smo) is known to inhibit the conversion of Ci from its active to repressive form. Knockdown of smo, which would be expected to increase levels of the repressive form of Ci, causes reduced growth (supplementary material Fig. S3B), suggesting that loss of the repressive Ci form is responsible for the overgrowth observed in ci mosaics. Knockdown of shaven (sv) (Fu et al., 1998) caused a reduction in the relative representation of mutant tissue, as well as abnormalities in patterning as assessed by irregularities in the arrangement of mutant ommatidia (Fig. 3F). Knockdown of pleiohomeotic (pho), a member of the Polycomb group of genes (Simon et al., 1992), resulted in a decrease in the representation of mutant tissue but also caused the partial transformation of antennae to legs (Fig. 3G).

Thirty-seven lines had no, or very small numbers, of mutant cells left in the adult eye, consistent with their elimination at some point in development. When expressed uniformly in the eye, the majority of these lines caused lethality or small eyes/heads (Fig. 3C), suggesting that these genes are necessary for cell proliferation or viability. For example, clones with reduced rolled (rl) function, which is the Drosophila ortholog of MAP kinase (Biggs et al., 1994), were absent and the overall size of the eye was reduced (Fig. 3H). Uniform knockdown of rl results in near-complete loss of the eye (Fig. 3K). This contrasts dramatically with the phenotypes elicited by knockdown of geminin, which encodes a regulator of the initiation of DNA replication (Quinn et al., 2001). Knockdown of geminin in clones resulted in the complete elimination of mutant cells (Fig. 3I; supplementary material Fig. S3C,D). However, uniform knockdown had no effect on the size or morphology of the adult eye or head (Fig. 3L). This suggests that geminin^RNAi cells are predisposed to elimination only in the presence of wild-type tissue, which is a characteristic of the phenomenon of cell competition (Morata and Ripoll, 1975). Additionally, uniform knockdown of geminin in the wing (supplementary material Fig. S3E,F) or ubiquitously (Fig. 3N) does not cause obvious morphological or size defects (supplementary material Fig. S3E). Three additional independent geminin^RNAi lines gave the same elimination phenotype when expressed in mosaic eyes, yet caused varying degrees of cell lethality when uniformly expressed in the eye, wing, or ubiquitously (supplementary material Fig. S3F), indicating that there might be a threshold above which more geminin knockdown causes autonomous cell viability defects. Thus, the CoinFLP-Gal4 system can identify novel regulators of cell competition.

The CoinFLP-Gal4 system enables the expression of RNAi constructs or cDNAs in subsets of clones. This allows the study of interactions between mutant clones and wild-type cells. However, to enable the study of interactions between two different types of mutant cells we developed a second system, CoinFLP-LexGAD/Gal4 (Fig. 4A). In this system, FLP recombination of canonical FRT sites results in excision of the STOP cassette and expression of the transcriptional activator LexGAD (Lai and Lee, 2006; Szuts and Bienz, 2000), which can activate the expression of genes that are downstream of the LexA operator (LexAop). Recombination between FRT3 sites excises both the STOP cassette and the LexGAD gene, resulting in the expression of Gal4. This results in the generation of mosaic tissue in which some patches express Gal4 and others express LexGAD.

Fluorescent reporters such as UAS-RFP and LexAop-GFP can separately mark the Gal4-expressing and LexGAD-expressing cells. When ey-FLP was used (crossing scheme in Fig. 4B), both Gal4-expressing and LexGAD-expressing clones were generated in the eye-antennal disc (Fig. 4C). As expected, clonal boundaries sometimes fell within individual ommatidia (Fig. 4C′) (Ready et al., 1976), generating ommatidia in which some cells expressed LexGAD while others expressed Gal4. If no recombination had occurred, the cells would express neither Gal4 nor LexGAD. Such clones were rare in the eye-antennal disc (Fig. 4C,C′) and the adult eye (supplementary material Fig. S4A-A″). When senseless (sens)-FLP was used, which is expressed in R8 photoreceptors (Chen et al., 2014), single cells expressed either Gal4 or LexGAD (Fig. 4D,D′). A number of Sens-positive cells expressed neither Gal4 nor LexGAD (Fig. 4D′). Interestingly, immediately posterior to the morphogenetic furrow, where Sens is first expressed, R8 cells do not express Gal4 or LexGAD. In more posterior rows of R8 cells, both Gal4-expressing and LexGAD-expressing cells are observed, indicating that the probability of conversion to a Gal4-expressing or LexGAD-expressing state increases with time of FLP expression. This contrasts with the almost complete conversion observed in the third instar disc with ey-FLP, which might be the result of higher expression levels than sens-FLP. Alternatively, because ey is expressed in the embryonic precursor cells of the eye before the onset of their proliferation in the larva, complete conversion might have occurred before proliferation begins. We also found rare cells expressing Gal4 or LexGAD that do not express Sens, immediately adjacent to Sens-expressing cells. These are likely to be cells of the R8 equivalence group, which express Sens transiently before the selection of a single Sens-expressing R8 cell (Frankfort et al., 2001). Using a heat shock-inducible source of FLP (hs-FLP), Gal4-expressing and LexGAD-expressing clones were observed in eye-antennal discs (Fig. 4E,E′), as well as patches that expressed neither.

To quantify recombination choice in the CoinFLP-LexGAD/Gal4 construct, we measured the number of Gal4-expressing and LexGAD-expressing cells in eye-antennal discs using ey-FLP and sens-FLP. For ey-FLP, we measured the areas of clones and found that Gal4-expressing clones are less frequent than LexGAD-expressing clones (19% compared with 80%, respectively; N=20 discs; s.d.=8%) (Fig. 4F). Clones expressing neither represent 1% of the total. For sens-FLP, because of ongoing recombination in the larval disc, we only counted the number of cells expressing Gal4 or LexGAD and again found that Gal4 expression was less frequent (8% compared with 92%, respectively; N=1311 cells). One interpretation could be that Gal4 and LexGAD cells are produced in similar quantities initially, but Gal4-expressing cells are disproportionally lost due to the toxicity of Gal4, as previously shown in other contexts (Rezaval et al., 2007). An alternative interpretation is that the relative recombination efficiency of FRT sites is less than for FRT3 sites in the CoinFLP-LexGAD/Gal4 construct. Since FRT and FRT3 sites have similar recombination frequencies (Schlake and Bode, 1994), the shorter distance between FRTs (411 bp) compared with FRT3s (3531 bp) in this construct might explain the difference in the frequency of LexGAD and Gal4 expression. Indeed, FRT recombination frequency is known to be reduced with increasing distance (Ringrose et al., 1999).

To demonstrate the wider applicability of the CoinFLP-LexGAD/Gal4 system, we examined other tissues and cell types and tested additional FLP sources. Using hs-FLP, we observed cells expressing either LexGAD or Gal4 in virtually every tissue examined at larval and adult stages, such as the wing disc (Fig. 4G; supplementary material Fig. S4B-B‴) and polyploid fat body cells (Fig. 4H; supplementary material Fig. S4C-C‴). Interestingly, in these polyploid cells, co-expression of LexGAD and Gal4 was frequently observed, presumably because the presence of multiple copies of the construct in the same cell allows both FRT- and FRT3-mediated recombination. We also imaged neurons expressing either LexGAD or Gal4, facilitated by the use of membrane-bound fluorescent markers to visualize fine projections such as axons and dendrites. sens-FLP mostly labeled R8 photoreceptors and their axons, which project toward the optic lobe of the larval brain (Fig. 4I), whereas GMR-FLP (Chen et al., 2014) labeled axons of the Bolwig nerve (Fig. 4J), and 20C11-FLP (Chen et al., 2014), which drives FLP from an Octβ1R regulatory fragment, labeled a small subset of neurons in the larval brain (supplementary material Fig. S4D). Importantly, whereas previous techniques such as Brainbow have labeled multiple cells with fluorescent markers, the CoinFLP-LexGAD/Gal4 system allows the inclusion of RNAi or overexpression transgenes to affect gene activity in labeled cells.

The CoinFLP-LexGAD/Gal4 system also allows the visualization of clonal boundaries by combining it with the GFP reconstitution across synaptic partners (GRASP) system (Feinberg et al., 2008; Gordon and Scott, 2009). In the GRASP system, two complementary parts of a ‘split GFP’ (spGFP1-10 and spGFP11) are fused to the extracellular domains of mouse CD4, one under UAS control and the other under LexAop control. Thus, the two components can reconstitute GFP at the boundaries where LexGAD-expressing cells contact Gal4-expressing cells (Fig. 5A). We illustrated the utility of the system in visualizing follicle cell clones in adult ovaries. Using hs-FLP, we generated mosaic egg chambers that expressed patches of Gal4 and LexGAD (crossing scheme in Fig. 5B). Most egg chambers are composed of the progeny of two follicle cell stem cells (Nystul and Spradling, 2010). However, in the example shown in Fig. 5C, we observed a stage 9 egg chamber that is likely to consist of three clones: two that express either Gal4 or LexGAD and a third that expresses neither. As predicted, GFP fluorescence was mostly observed at the interface between LexGAD-expressing cells and Gal4-expressing cells (Fig. 5C,C′). Interestingly, a low level of punctate GFP expression was observed far from the boundary, especially in the part of the egg chamber that expresses UAS-spGFP1-10 (Fig. 5C′). The reason for this is unclear, although the puncta were aligned in circumferential bands that would be consistent with their secretion into the basement membrane by follicle cells that are crawling circumferentially around the egg chamber (Haigo and Bilder, 2011).

We also visualized clonal boundaries in the wing imaginal disc using GRASP. When clones were generated with hs-FLP, we observed both LexGAD-expressing and Gal4-expressing cells, as well as GFP fluorescence at the boundaries between them (Fig. 5D-F″). In a confocal section midway between the apical and basal surfaces of the disc epithelial cells, GFP was observed at the clonal boundary as well as in puncta on both sides of the boundary (Fig. 5D,D′). These puncta might be the contacts of lateral protrusions (Demontis and Dahmann, 2007). Reconstructed confocal sections of the basal surface (Fig. 5E,E′) revealed thin processes reminiscent of cytonemes (Hsiung et al., 2005), which were observed emanating from the clonal boundary and aligned with the dorsoventral axis (Fig. 5E). GFP puncta were sometimes observed at the tip of these processes (Fig. 5E′), suggesting that they make more stable contacts at their endpoints. A 3D reconstruction shows the extent of cell-cell contact along the lateral surface of wing disc cells (Fig. 5F-F″; supplementary material Movie 1). Surprisingly, these images reveal less cell-cell contact near the apical surface as compared with the basal surface (Fig. 5F′,F″). Furthermore, we observed points of contact between the apical membrane of cells in the disc proper and the overlying peripodial membrane (Fig. 5F′) (Demontis and Dahmann, 2007).

Many genetic changes are known to cause mutant clones to adopt rounder shapes, presumably because of altered adhesion between mutant and wild-type cells. For example, activation of Dpp signaling through the Thickveins (Tkv) receptor causes increased growth of cells and smoother clonal boundaries with wild-type cells (Martin-Castellanos and Edgar, 2002). Using GRASP, we find that, whereas wild-type clones have irregular boundaries (Fig. 5G), clones expressing an activated Tkv receptor (Tkv^Q253D) have smooth boundaries (Fig. 5H). Thus, the CoinFLP-LexGAD/Gal4 system together with GRASP can be used to visualize boundaries and regions of cell-cell contact between genetically distinct populations of cells.

The CoinFLP-Gal4 system facilitates mosaic Gal4/UAS screens by automatically generating mosaic tissue in a developmentally restricted manner. We have used this system to perform a pilot RNAi screen, focusing on genes that are refractory to traditional mosaic analysis. We identified several mutant phenotypes, one of which, induced by geminin^RNAi, appears to cause mutant cells to be eliminated by cell competition. geminin has not been previously implicated in cell competition and therefore deserves further study. Importantly, this phenotype could only be identified in the context of a mosaic environment, suggesting that the CoinFLP-Gal4 system can be used for larger scale screens to find novel regulators of cell-cell interactions.

Many of the shortcomings of mutagenesis-based genetic screens have been overcome by screens that use RNAi. There are collections of fly stocks available that target most, if not all, annotated genes (Dietzl et al., 2007; Ni et al., 2011). With RNAi screens, the identity of the gene is known from the outset, thus saving the effort required for gene identification by mapping and/or genome sequencing. Additionally, targeting the expression of these RNAi transgenes to a specific tissue (e.g. the eye) overcomes any requirement for the genes being tested at earlier stages of development or in all tissues. However, until now, it was not easy to conduct mosaic screens using RNAi in which a reliable ratio of wild-type and mutant cells is consistently achieved, such that deviations from the normal ratio can be easily scored. Additionally, CoinFLP-Gal4 can be used to screen mosaic phenotypes associated with the overexpression of genes using collections of cDNAs expressed under the control of UAS elements (Bischof et al., 2013).

The CoinFLP-LexGAD/Gal4 system allows the control of gene expression in two different populations within the same tissue. For example, this technique can be used to generate different labeled mutant clones and observe their interaction. In addition, we have used this system with GRASP to visualize clonal boundaries in the egg chamber and wing imaginal disc, especially the surfaces between adjacent cells and other points of contact (e.g. along processes). There is some controversy as to whether GRASP increases the affinity of cell-cell interactions (Feinberg et al., 2008; Kim et al., 2012). It is likely that future versions of GRASP will address some of these possible shortcomings. The generation of other split proteins that can be reconstituted to provide an activity that can be scored in adult tissues might further enhance the ability to use CoinFLP-LexGAD/Gal4 in live undissected flies. Finally, CoinFLP-LexGAD/Gal4 could be used to perform more elaborate mosaic screens to study cell-cell interactions, where gene activity is altered in one population of cells while the other cell population is systematically screened by RNAi or overexpression libraries.

In its present form, the CoinFLP system is useful for inducing Gal4 or LexGAD expression from a constitutive promoter in mosaic patches in a specific tissue. However, further modifications to this system can be made. For example, instead of the Actin 5C promoter, a tissue-specific promoter could be used to help spatially restrict gene expression. Similarly, different tissue-specific sources of FLP can be used to restrict mosaicism to a desired cell lineage. Furthermore, other genes can be used in place of Gal4 or LexGAD, such as QF, which drives genes under QUAS control (Potter et al., 2010). Also, different arrangements of variant FRT sites can be used, such as employed in the FlEx system (Schnutgen et al., 2005), or additional variant FRT sites such as FRT5 (Schlake and Bode, 1994) can be added to increase the number of unique mosaic patches. Finally, one can manipulate the relative quantity of Gal4-expressing and LexGAD-expressing cells by swapping the two coding sequences, such that Gal4 may then be more frequent than LexGAD, or by changing the distance between FRT sites.

Detailed descriptions of how the CoinFLP systems were constructed are provided in the supplementary Materials and Methods, including plasmid maps of CoinFLP-Gal4 (supplementary material Fig.S5) and CoinFLP-LexGAD/Gal4 (supplementary material Fig.S6).

All crosses were maintained at 25°C on standard fly medium (BDSC) supplemented with yeast pellets. Most fly lines used are available from the Bloomington Drosophila Stock Center at Indiana University (BL) unless otherwise noted. Fly lines used: ey-FLP (X, BL5580), Ubx-FLP (BL42718), UAS-white^RNAi (III, BL33623), Act>y+>Gal4 (II, BL3953), UAS-EGFP (II, BL6874), UAS-dcr2 (X, BL24646), 40AFRT (Xu and Rubin, 1993), ey-FLP; 40AFRT ubi-GFP (derived from BL5189), UAS-RpL24^RNAi (BL34569), UAS-ex^RNAi (BL28703), y v; attP2 (negative control for TRiP stocks, BL36303), UAS-dcr2; nb-Gal4 (BL25754), UAS-dcr2; Act-Gal4 (BL25708), UAS-ECFP-β-actin (BL7064), 10XUAS-IVS-mCD8::RFP, 13XLexAop2-mCD8::GFP (BL32229), LexAop-rCD2::RFP (II) (Awasaki et al., 2011), LexAop-rCD2::GFP (Lai and Lee, 2006), UAS-His2A-RFP (III) (Emery et al., 2005), UAS-CD4::spGFP1-10 (III), LexAop-CD4::spGFP11 (III) (Gordon and Scott, 2009), UAS-ci^RNAi (BL28984), UAS-sv^RNAi (BL27269), UAS-pho^RNAi (BL31609), UAS-rl^RNAi (BL34855), UAS-geminin^RNAi (GLV21038, BL35673, used in Fig. 3), sens-FLP (BL55768), GMR-FLP (Chen et al., 2014), UAS-tkv^Q253D (BL36536), UAS-ebony^RNAi (BL28612), UAS-forked^RNAi (BL41687), UAS-mwh^RNAi (BL34862), UAS-scrib^RNAi (VDRC 27424), UAS-rpr (BL5824), UAS-myc (Johnston et al., 1999), UAS-smo^RNAi (BL53348), 20C11-FLP (BL55766).

Stocks used in experiments shown in the supplementary material that are not described in main text: UAS-geminin^RNAi GL01299 (BL41869), UAS-geminin^RNAi (BL30929), UAS-geminin^RNAi HMS02954 (BL50720). ey-FLP stocks used are listed in supplementary material Table S1 and UAS-RNAi stocks used are listed in supplementary material Table S2. Supplementary material Table S3 includes descriptions of CoinFLP stocks deposited at the Bloomington Drosophila Stock Center. Information on how we selected the RNAi lines for the CoinFLP-Gal4 screen is described in the supplementary Materials and Methods.

To aid in testing multiple UAS lines by genetic crosses, such as conducting a genetic screen, we generated a ‘tester’ fly stock that contains several transgenes to produce mosaic adult eyes with white^RNAi marking Gal4 patches: ey-FLP, UAS-dcr2; CoinFLP-Gal4 attP40; UAS-white^RNAi. This tester stock breaks down relatively often and we therefore recommend that this stock is constructed each time from stocks that we have deposited in the Bloomington Drosophila Stock Center. Additionally, as a comparison for mosaic Gal4 expression, we constructed an analogous tester stock that induces uniform expression of Gal4 from the Actin 5C promoter in most, if not all, eye-antennal disc cells following FLP-induced excision of a STOP cassette: ey-FLP, UAS-dcr2; Act>y+>Gal4 UAS-GFP.

Imaginal discs were dissected from wandering third instar larvae, or ovaries were dissected from adult flies, fixed in 4% paraformaldehyde for 20 min, and washed with PBS. Samples were incubated in DAPI and mounted on slides with SlowFade Gold mounting medium (Life Technologies). Fixed samples processed for antibody staining were permeabilized in PBS+0.1% Triton-X100 for 20 min, blocked with PBS+0.1% Triton-X100+5% normal goat serum for 1 h, and incubated with primary antibodies diluted in blocking solution overnight at 4°C. Samples were washed three times for 15 min each in PBS+0.1% Triton-X100. Samples were incubated with secondary antibodies and washed as for primary antibodies. Mounted samples were imaged on a Zeiss 700 confocal microscope. Confocal slices were processed with ImageJ software (NIH). Confocal stacks were processed with Volocity software (Perkin Elmer) to generate 3D projections.

Staining reagents: guinea pig anti-Sens (1:500; H. Bellen, Baylor College, Houston, USA); rabbit anti-GFP (1:800; Abcam ab290), which recognizes spGFP1-10 and reconstituted spGFP, but not spGFP11 (Gordon and Scott, 2009); 1 ng/μl DAPI.

For experiments requiring heat shock induction of the hs-FLP transgene in larvae, 48 h after egg deposition (AED) larvae were placed in a 37°C water bath for 30 min. Larvae were dissected as wandering third instar larvae at 120 h AED. For heat shock induction in follicle cells, adult female flies were heat shocked for 10 min and ovaries were dissected 2 days later.

Adult female flies were frozen overnight and eyes were imaged on a Leica transmitted light microscope (TL RCI) at identical settings. White versus red ratios in mosaic eyes were determined by measuring the area (pixels²) of red eye tissue in Photoshop CS6 (Adobe) and comparing with the total area of the eye. To image fluorescence in adult eyes, flies were sedated with CO₂ and Flynap (173010, Carolina) and fluorescence was imaged using a Zeiss stereo microscope (Lumar.V12).

Wings were mounted onto slides using Canadian Balsam medium (Gary's Magic Mount) and imaged on a Leica TL RCI. Wing area was measured in ImageJ using the polygon selection tool. Statistical significance between groups was determined by one-way ANOVA using Tukey's test. At least 30 wings/genotype were used.

Confocal images of eye-antennal discs were processed using ImageJ. Clone area and total disc size were measured using the polygon selection tool. Cell number was measured using the multi-point selection tool. Statistical significance between groups was determined by one-way ANOVA using Tukey's test. For quantifications of CoinFLP-Gal4 recombination with ey-FLP, ten eye-antennal discs were used for each genotype (ey-FLP UAS-dcr2/+; CoinFLP-Gal4 UAS-GFP/+; UAS-X/+). For quantifications of CoinFLP-LexGAD/Gal4 with ey-FLP, 20 eye-antennal discs were used (genotype: ey-FLP/UAS-CD8-RFP, LexAop-CD8-GFP; CoinFLP-LexGAD|Gal4/+). For quantifications of CoinFLP-LexGAD/Gal4 with sens-FLP, 1311 cells were counted (genotype: sens-FLP/UAS-CD8-RFP, LexAop-CD8-GFP; CoinFLP-LexGAD|Gal4/+).

We thank Taryn Sumabat for discussions; Joanna Downes for help with plasmid constructions; Cassandra Van and Geoffrey Logan for experimental assistance; Laura Johnston, Tzumin Lee, Duojia Pan and Kristin Scott for fly stocks; Gerry Rubin, David Strutt and Larry Zipursky for plasmids; and Joelle Lomax for suggesting the name ‘CoinFLP’.