A versatile platform for creating a comprehensive UAS-ORFeome library in Drosophila

Traditionally, genetic analysis in Drosophila has relied on loss-of-function approaches to determine the specific function of genes. Such mutagenesis screens have contributed substantially to our understanding of gene functions and biological processes. Over the years, this strategy was extended and recently culminated in the creation of a genome-wide RNAi library for targeted knockdown of genes in vivo (Dietzl et al., 2007) (other similar resources are ‘TRiP’ and ‘NIG-Fly’, see www.flyrnai.org and www.shigen.nig.ac.jp/fly/nigfly, respectively). Despite their merits, loss-of-function studies have failed to reveal the function of the majority of all Drosophila genes, as they show no obvious loss-of-function phenotype (Miklos and Rubin, 1996). Some of this is likely to be due to functional redundancy, which could be addressed by, for example, combinatorial RNAi strategies. A likely reason for the frequent ‘absence’ of loss-of-function phenotypes is the use of assays that are simply not sensitive enough to detect subtle defects (e.g. Ramani et al., 2012). Thus, the function of the majority of all Drosophila genes cannot be identified by conventional loss-of-function or knockdown screens, and alternative approaches, such as the use of sensitised backgrounds, as well as an increase in assay sensitivity are required to mitigate this limitation.

By contrast, extensive mis- and overexpression screens in Drosophila are less common, though they can be equally informative for elucidating gene functions (e.g. Wittwer et al., 2005; Prelich, 2012). So far, such screens have almost exclusively relied on random transposon insertions containing an enhancer-promoter (EP) element that can be used to drive the expression of genes flanking the insertions (Rørth, 1996; Bellen et al., 2011). In another case, a collection of transgenic lines generated from a specific set of heterologous open reading frame (ORF) constructs was used for misexpression (Xu et al., 2008; for a review, see Zhong and Yedvobnick, 2009). As with the RNAi library, the binary Gal4/upstream activating site (UAS) system (Brand and Perrimon, 1993) is used in these gain-of-function screens to achieve spatiotemporal control of gene expression; naturally, these screens can also be performed in sensitised backgrounds. Screens utilising such random insertion lines, however, are not saturating, an inherent limitation of using transposons. Furthermore, unambiguous identification of the phenotype-causing ‘event’, which does not necessarily have to be the activation of a neighbouring gene, often causes considerable complications and efforts.

Thus, a more systematic and controlled method of misexpression is required to overcome these shortcomings. In particular, the new strategy should potentially allow saturation and any resulting phenotype should be precisely linked to a specific gene. We propose the generation of a transgenic ORFeome library originating from specifically designed overexpression constructs. Up to now, however, the in vivo use of even small sets of UAS-ORF constructs in Drosophila has been very limited (Xu et al., 2008). The considerable efforts required for cloning and sequencing full-length expression constructs and for establishing and maintaining the large number of resulting transgenic lines has hampered implementation of such a strategy. The recent development of the site-specific ΦC31 integrase system in Drosophila (Groth et al., 2004; Bateman et al., 2006; Venken et al., 2006; Bischof et al., 2007) overcomes some of these challenges. In particular, it is now feasible to inject large collections of constructs, as this method guarantees efficient integration. Targeted insertion eliminates the need for mapping, and selection of a suitable attP (phage attachment) site minimises unpredictable position effects. Consequently, the effects of transgenes can be compared more reliably. A further advantage of the ΦC31 integrase system is that lost transgenic lines can be easily and identically reconstructed.

Here, we present a strategy for establishing a genome-wide UAS-ORF in vivo library, exemplified by the generation of a pilot library containing 1149 fly lines. Site-specific integration and the use of molecular barcodes facilitate the efficient generation and identification of large numbers of transgenic flies. Furthermore, we investigated the effects of epitope tags on transgene functionality. We present a system that enables modification of the ORFeome library in vivo, including the exchange of promoter regions and epitope tags simply by crossing appropriate fly strains. In summary, this library provides a powerful tool for Drosophila genetics and offers a rapid screening method for identifying novel regulators or effectors of genetic pathways.

First, pUAST.attB (Bischof et al., 2007) was digested with NheI-HindIII, releasing the loxP site fragment, which was replaced by a duplex oligonucleotide containing the SwaI site (NheI and HindIII sites were destroyed). Next, the UAS-hsp70 promoter fragment was amplified from pUAST.attB and cloned into the plasmid using SwaI and KpnI. The SV40 trailer (cleaved with KpnI-BamHI) was replaced with a duplex oligonucleotide containing the following restriction sites: KpnI, NheI, XhoI, HindIII and a BamHI compatible overhang (destroying the BamHI site after insertion). Next, we digested the plasmid with NheI-XhoI and inserted a tubulin 3′UTR, PCR-amplified from pM{3xP3-RFPattP} (Bischof et al., 2007). We digested this clone with KpnI, followed by blunting, and then with NheI to insert a Gateway cassette, either without tag (pUASg.attB) or with the 3xHA tag (pUASg-HA.attB). The cassette for pUASg.attB was released from the Gateway plasmid pAW by SacI, followed by blunting, and then NheI digestion. The cassette for the pUASg-HA.attB plasmid was released from the Gateway plasmid pAWH by EcoRV and NheI digest. Details of plasmids pAW and pAWH can be found at the Drosophila Genomics Resource Center (DGRC) vector collection (https://dgrc.cgb.indiana.edu/vectors/store/vectors.html).

We digested pUASg.attB with BglII-AgeI and inserted a duplex oligonucleotide containing the shortened and mutated FRT5 variant (CGAAGTTCCTATTCTTCAAAAGGTATAGGAACTTCA) (Schlake and Bode, 1994).

Then we digested with AgeI-NotI and inserted the 5′ part of the Gateway cassette, amplified from pUASg.attB. Next, the plasmid was digested with XbaI-NheI, and both the 3′ part of the Gateway cassette (amplified from pUASg.attB, pre-digested with XbaI-KpnI) and the 3xHA stretch (amplified from pUASg-HA.attB, pre-digested with KpnI-NheI) were simultaneously inserted.

Next, we cleaved with SwaI-BglII and inserted a loxP-UAS-hsp70 fragment amplified from pUASg.attB (loxP sequence was inserted by the forward primer). Finally, we cleaved the plasmid with KpnI and inserted an FRT2 oligonucleotide (CGAAGTTCCTATTCTCTACTTAGTATAGGAACTTC) (Schlake and Bode, 1994).

This is our current destination vector used for the further cloning of the ORFeome library. This plasmid differs from pTF-HA.attB in that the sequence between the FRT2 and the 3xHA tag was replaced with a 24-aa-long linker region (FL sequence: ILGAPSGGGATAGAGGAGGPAGLI) that is thought to minimise steric interference between the native protein and the epitope tag (Jankovics and Brunner, 2006). We cleaved pTF-HA.attB with KpnI-NheI and inserted the synthesised fragment FRT2-FL-3xHA via these sites, giving rise to pGW-HA.attB.

We digested pUAST.attB (Bischof et al., 2007) with BamHI and removed the UAS-hsp70-SV40 cassette. This cassette was replaced by a duplex oligonucleotide containing a multiple cloning site.

The MCS (multiple cloning site) of the pattB plasmid was deleted by NheI-XbaI and replaced with the linker NheI-MluI-AvrII-XbaI (NheI site being destroyed), followed by DraII digest and insertion of a oligonucleotide containing the sites DraII-XhoI-NheI-NotI-KpnI-BglII. The BglII site was used to insert the mutant FRT5 variant. The plasmid was opened with NheI-KpnI, made blunt, and used to insert a blunted lexA operator (lexO) BamHI-EcoRI fragment from pLOT (Lai and Lee, 2006). Finally, a yellow marker gene released with SalI from a flp-out cassette (Basler and Struhl, 1994) was inserted into the compatible XhoI site, giving rise to pPSlexO.attB.

We deleted the white gene and the loxP site from pattB by DraII-NheI digestion, blunted, re-ligated, digested with BamHI-XhoI and used the plasmid in a four-fragment ligation together with the following fragments: BamHI-FRT2-HindIII, HindIII-eGFP-NotI and NotI-tubulin 3′UTR-XhoI. The resulting plasmid from this ligation was then digested with XhoI to insert a yellow gene (identical to pPSlexO.attB), generating pTSeGFP.attB.

The VNm9 and VC155 fragments were PCR-amplified from plasmids VNm9 and VC155 in pCS2 (Saka et al., 2007), introducing the flanking restriction sites HindIII and NotI. Additionally, a myc tag was introduced at the 5′ end of VNm9. Plasmid pTSeGFP.attB (without the yellow insertion) was digested with HindIII-NotI, followed by insertion of VNm9 and VC155, respectively. Finally, the yellow marker was inserted as already described.

This plasmid is identical to p3xP3-eGFP/vas-ΦC31(+/−NLS)attB in Bischof et al. (Bischof et al., 2007), except for dΦC31 replacing Streptomyces phage ΦC31. dΦC31 is a Drosophila codon-optimised ΦC31 integrase, differing in 172 nucleotides from the phage integrase ORF.

Detailed information on the construction of the reporter plasmids placZ.attB, placZ-2.attB, pEGFP.attB and pEGFP-2.attB and on the FRT test constructs is available upon request.

Molecular barcoding using randomised duplex oligonucleotides was used to facilitate identifications of plasmids and fly lines. We first isolated a XhoI-HindIII flanked lacZ gene as a stuffer fragment and inserted it into the XhoI-HindIII site of the various vectors to be barcoded. Plasmids were then digested and the stuffer replaced with the barcode oligonucleotides. For the pilot library we used the following barcode design: 5′-XhoI-ANNTANNNATNNNNTAANNNNTANNNATNNNTANNG-HindIII-3′, i.e. 36 nucleotides of which 21 are randomised. The 5′-phosphorylated barcode oligonucleotide mixtures were annealed according to standard protocols and the restriction sites were created as ‘sticky-end’ overhangs. The barcodes were ligated into the plasmids in a 250 μl ligation mix overnight at 16°C. This we next used to transform 50 tubes of 50 μl aliquots of DB3.1 cells with 2.5 μl ligation mix each; these reactions were heat-shocked for 45 seconds, then combined and added to 100 ml SOC medium. After 1 hour shaking at 37°C, 100 μl aliquots were plated on LB+Amp plates to calculate the diversity of the barcodes. We scored ~450 colonies, which results into an actual diversity of 450,000 (dilution 1:1000). 2xTY medium (300 ml) containing 75 μg/ml ampicillin and 10 μg/ml chloramphenicol was added to the remaining culture, which was grown for 19 hours at 37°C. Plasmids were purified using Qiagen Maxi Prep Kit.

The barcode used for the latest version of our destination vectors, pGW-HA.attB, has the following composition: 5′-XhoI-ANNTGNNNACNNNNTGANNNNACNNNATNNNGANNG-HindIII-3′. The main difference from the previous version is an increase in the GC content.

Full-length ORFs were cloned from the Drosophila gene collections releases 1-3 or from the Schneider cell SD-pOT2 cDNA library (Berkeley Drosophila Genome Project) with gene-specific primers using a two-step strategy in which full-length attB1 and attB2 sites are introduced into the PCR product as described in the Gateway Cloning manual (Invitrogen). The forward primer had a sequence AAAAAGCAGGCTTCAAC before methionine codon and gene-specific sequence [sequence corresponding to Drosophila Kozak consensus CAA(A/C) underlined] (Cavener, 1987). Two versions of the reverse oligonucleotide were used, one with native stop codon and one without stop, with AGAAAGCTGGGTC flanking sequence. The inserts were cloned to pDNR221 vector by BP recombination reaction and plated on kanamycin-containing LB agar plates. Single clones were picked into 96-well deep well plates containing 1.3 ml Terrific Broth medium and cultured for 22-24 hours at 37°C. Miniprep DNA was purified using the Promega Wizard SV96 Kit. Correct recombinants were identified by end-sequencing with M13F and M13R primers.

For the entry clones for which full-length ORF sequence could not be obtained by end sequencing, we amplified the inserts by PCR using attB1 and attB2 primers. The PCR products were pooled, nebulised and sequenced using a Roche 454 Sequencer at the Institute of Biotechnology, University of Helsinki.

The entry clones were transferred to barcoded pUASg.attB (for stop-containing clones) and pUASg-HA.attB (for ORFs without stop codon) destination vectors by a Gateway LR recombination reaction and selected on ampicillin.

Single ORF clones were arrayed in 96-well plates and the plasmid concentration normalised to 100 ng/μl. To maximise the recovery of the transgenes in the pool injections, the ORF clones were arrayed according to their size over several multiwell plates, i.e. the clones within one multiwell plate span the smallest possible size range. Before pooling, the barcodes were sequenced from the expression clones to associate the barcode information with the individual ORF.

The ORF clones were pooled with multi-channel pipettes and the mixes were cleaned using Midiprep Qiagen-tips. For injections, we diluted the pools to a concentration of 100 ng/μl.

We inserted a Drosophila codon-optimised ΦC31 integrase construct, p3xP3-eGFP/vas-dΦC31attB, into the attP landing site ZH-attP-2A (Bischof et al., 2007). The resulting transgenic line, vas-dΦ-zh2A, is doubly fluorescence-marked with RFP and eGFP. Both markers are under the control of the 3xP3 promoter (Horn et al., 2000). The RFP marker originates from the attP landing site construct pM{3xP3-RFPattP}, the eGFP from the integrase construct. To create the line ΦX-86Fb, we combined the vasa-integrase transgene with the ZH-attP-86Fb landing site. The genotype of this line is y⁻w⁻ M{eGFP.vas-int.Dm}ZH-2A; +; M{3xP3-RFP.attP}ZH-86Fb. The line ΦX-86Fb is available at the Bloomington Drosophila Stock Center (Indiana University), along with other lines created in a similar fashion. We maintain these ΦC31 integrase containing stocks at 18°C.

PSlexO-86Fb, TSeGFP-86Fb, TSVNm9-86Fb, TSVC155-86Fb: The constructs pPSlexO.attB, pTSeGFP.attB, pTSVNm9.attB and pTSVC155.attB were injected into line ΦX-86Fb. Lines were made homozygous for these transgenes and combined with an X chromosome-linked hsp70-flp construct. The fly strains used for bimolecular fluorescence complementation (BiFC) analysis were combined with a GMR-Gal4 on the second chromosome. Other used driver lines were MS1096-GAL4, GMR-GAL4, ey-Gal4, dpp-GAL4, en-GAL4 and C765-Gal4.

We injected the plasmid DNA pools into ΦX-86Fb. As a rule of thumb we injected ~100 embryos per ten different constructs, i.e. for a pool of ~100 ORFs we injected ~1000 embryos (ten slides). The embryos were dechorionated, dried, covered with 10S Voltatef oil and injected with either a simple syringe-microinjection device or with an Eppendorf FemtoJet-TransferMan setup combined with a micromanipulator mounted on a Zeiss Axi inverted microscope. Glass capillaries were pulled on a Narishige PN-30 puller and opened on a grinder.

Injected embryos were kept in a moist chamber at 18°C for 2 days, then larvae were transferred to fly food vials and raised at 25°C (~80-120 per vial). Two G0 males were crossed to three to four y⁻w⁻ virgins per vial; G0 females were discarded. Single F1 transgenic progeny males were crossed to a third chromosome balancer line (y⁻w⁻; D gl3 / TM3 Sb Ser) to subsequently create a balanced stock. We often took two F1 transgenic males from the same vial and crossed them out individually, as we often acquired different transgenic lines from these males. Established balanced stocks were not subsequently monitored in order to create homozygous stocks.

Three days after setting up the F1 crosses, we subjected the males to single fly PCR procedures. PCR was carried out with a primer pair aligning to the tubulin 3′UTR of the vector and to the flanking genomic region of the 86Fb landing site (tub-F2 new: ATTTATGTGACTATGGTAGGTCG; 86Fb-Rev: GCTTAGCTTCTGGGTGCATGTGACCG). PCR products were sequenced to identify the barcode composition. This information permits association of the individual outcrosses with specific ORFs, from which we could decide which lines to keep and balance, and which to discard (discarded lines represent recurring transgenes).

We used a vial setup and tested several heat-shock regimes, guided by conditions used in Parks et al. (Parks et al., 2004). We crossed four males from the ORF library to approximately ten females carrying the swap construct and a hsp70-flp transgene (all flies homozygous for the constructs). At day 3, we subjected the progeny to a single 30-minute heat-shock. Offspring virgin females were collected and mated to y⁻w⁻ males: four females and three males per vial (F1 cross). The offspring of these crosses was screened for the appropriate marker combinations (see Results). We noticed that often only a few flies with the desired marker combinations were found per positive outcross. Note that in other experiments we outcrossed males instead of females in the F1 crosses and obtained similarly efficient swapping rates.

Protein extracts were obtained by boiling (5 minutes at 95°C) heads or third instar wing imaginal discs in 30 μl NuPAGE LDS sample buffer (Invitrogen). Protein extracts were centrifuged at 14,100 g for 5 minutes, separated on NuPAGE 4-12% Bis-Tris Gel (Invitrogen) and transferred onto Amersham Hybond-C Extra membrane (Amersham Biosciences). HA-tagged proteins were detected with mouse anti-HA antibody (HA.11, Covance, 1:5000); equal loading was estimated with mouse anti-α-tubulin antibody (DM1A, Sigma, 1:5000).

Antibodies used were mouse anti-HA antibody (HA.11, Covance, 1:500) and secondary goat anti-mouse Alexa594 (1:400, Molecular Probes). DAPI (0.4 μg/ml) was used for staining DNA. Fixation, immunohistochemistry and imaging were conducted via standard protocols.

RNA isolation, cDNA synthesis and single fly PCR were carried out according to standard protocols.

GenBank accession numbers are as follows: pUASg.attB, KC896836; pUASg-HA.attB, KC896837; pGW-HA.attB, KC896838; pattB, KC896839; placZ.attB, KC896840; placZ-2.attB, KC896841; pEGFP.attB, KC896842; pEGFP-2.attB, KC896843.

The transgenic UAS-ORF lines will be available from FlyORF (http://www.flyorf.ch).

We first constructed two Gateway-based destination vectors (Fig. 1B): one for the expression of full-length ORFs with their native stop codon (untagged clones, stop clones), and a second for ORFs that are fused to a C-terminal 3xHA (hemagglutinin) tag after the shuttling event (tagged clones, HA or 3xHA clones). Three tandem copies of the HA epitope tag were used to increase the sensitivity and signal-to-noise ratio in biochemical and histochemical assays (Jarvik and Telmer, 1998). These vectors are equipped with a partially randomised oligonucleotide of 36 bp for vector identification. Some positions within the barcode are kept constant to provide a recognisable signature, and the remaining positions are completely randomised. The used barcode has a theoretical diversity of 4²¹ (21 random positions, 15 fixed), allowing any transgene in a genome-wide collection to be unambiguously identified even in the presence of sequencing errors. In practice, the diversity is limited by the transformation efficiency of the barcode library. Our current version of the vector has ~450,000 independent clones (see Materials and methods). In subsequent steps, this molecular barcode serves as a unique identifier for any expression clone and the corresponding fly line. These vectors were used to generate a ‘pilot library’ of 655 different Drosophila genes, and two versions of this gene set were created: an untagged set and a 3xHA-tagged set. These genes were selected based on their loss-of-function phenotypes in Drosophila S2 cells (Björklund et al., 2006) or because they belong to pathways that have previously been implicated in growth, cell size regulation and cell cycle progression. All the ORFs were provided with a Drosophila Kozak consensus sequence to support efficient translation (see Materials and methods) (Cavener, 1987).

Transgenic constructs were inserted into the Drosophila genome at the cytological position 86F, using the previously generated attP insertion ZH-attP-86Fb (Bischof et al., 2007). This landing site was chosen based on a number of parameters that are crucial for the efficient generation of a large transgenic library and for its subsequent utilisation. We previously evaluated this line, together with others, based on the expression profile in wing discs (Bischof et al., 2007). Additionally, we quantified expression levels in 11 ZH-attP lines by measuring β-galactosidase activity upon expression of UAS-lacZ reporter transgenes by the C765-Gal4 driver. In these assays the site attP-86Fb provided one of the highest expression levels among the tested lines (supplementary material Fig. S1). In summary, ZH-attP-86Fb offers (1) a high integration rate, (2) strong transgene expression, (3) ease of scoring transgenic offspring owing to a moderately strong white marker expression, which enables differentiation of hetero- and homozygous transgenics, and (4) good overall fitness and high fertility.

Probably owing to a genomic location (supplementary material Fig. S2A) that favours strong transgene expression, the line also displays some background expression (detected during early embryonic development; not shown). We achieve a high homozygosity rate of ~90% with inserted transgenes at this attP site (based on 1107 UAS-ORF lines), further indicating that this site is well suited for library construction. The site attP-86Fb was combined with a Drosophila codon-optimised ΦC31 integrase (Bischof et al., 2007) located on the X chromosome, giving rise to the line ΦX-86Fb, which we used for creating the pilot library.

In addition to the library vectors mentioned above, we constructed multiple transformation vectors suitable for the ΦC31 integration method. These include the vector pattB for cloning of, for example, genomic rescue constructs, and the four reporter constructs placZ.attB, placZ-2.attB, pEGFP.attB and pEGFP-2.attB (supplementary material Fig. S2B; see also the FlyC31 website at www.flyc31.org). These vectors are perfectly suited to increase the versatility of the ΦC31 integrase system.

To avoid time-consuming and tedious single construct injections, we mixed ORF constructs together and injected them as pools into the ΦX-86Fb host (Fig. 1A). Advantages of the pooling strategy include: (1) specific transgene plasmid purifications for injections are reduced to one per pool instead of many individual purifications; (2) the exchanges of injection needles are drastically reduced, which saves considerable time; (3) injected embryos do not have to be kept separately, which simplifies the handling; and, most importantly, (4) pool injections reduce the number of embryos injected substantially as one round of injection leads to the recovery of many different transgenes versus only one in the case of single construct injections.

Initially, we tested several pool sizes of up to 400 plasmids without seeing a specific effect of pool sizes on transgene recovery rates (supplementary material Table S1). Based on practical considerations, we continued to use pools of ~100 constructs for injections (‘96-multiwell format’). Such a pool size led to a recovery of ~60% of the included ORFs from 100-120 outcrosses per injection round (supplementary material Table S1). The individual plasmid concentrations are normalised in the pool to maximise the recovery of all transgenes. The transgenic progeny is determined by single fly PCR, i.e. amplifying the plasmid region containing the barcode, followed by Sanger sequencing. Standard vector/86Fb-specific PCR primer pairs simultaneously identify the barcode and confirm the site-specific integration into the 86Fb site (Fig. 1A). The previous assignment of an individual barcode to each ORF uniquely identifies the transgenic lines without needing to sequence into the actual ORFs. Each new line that we obtained was balanced, whereas repeatedly occurring lines were discarded. Constructs that were not recovered in the transgenic offspring were pooled again and injected in a subsequent round. Altogether, we created transgenic lines for 547 untagged and 602 HA-tagged ORFs (supplementary material Table S2).

A tagged library holds major advantages over an untagged library: (1) a single antibody can be used to detect any ORF; (2) cross-reaction with related proteins can be avoided, as an antibody specific to the tag can be used; (3) the tagged protein can be distinguished from the endogenous, untagged protein; and (4) immunochemistry becomes possible for even poorly immunogenic proteins or proteins that lack a specific antibody (Jarvik and Telmer, 1998). However, because epitope tagging can also interfere with protein function, we examined the potential of the C-terminal 3xHA tag to alter protein function in overexpression experiments. We crossed all UAS-ORF lines that we had obtained in both versions to the wing-specific MS1096-Gal4 driver line and recorded the phenotypes. The crosses were scored blind. We classified the effects on a scale from 0 (no effect) to 4 (lethal). Categories 1-3 represent different phenotypic strengths with 1 indicating mostly mild patterning or growth defects, 2 indicating severe growth defects, and 3 indicating complete or almost complete absence of wings. Overall, we could test 473 genes in both versions. Of these, 230 showed no phenotype in either version and were therefore not informative for this purpose. We also excluded 46 genes that showed a weak effect (category 1) in the untagged version but not the tagged version and 22 with the opposite behaviour, as they could have been false negatives due to slight experimental alterations, such as incubation temperature. Of the remaining 175 transgenes, 20 showed a strong phenotype (>2) in the tagged version but no effect in the untagged version (11.4%, ‘false positive’). In addition, 38 showed a strong phenotype (>2) in the untagged version but no effect in the 3xHA version (21.7%, ‘false negative’). The remaining genes induced similar phenotypes that sometimes varied in phenotypic strength by no more than one category (11 cases showed a divergence by two categories).

We also used the 3xHA epitope tag to monitor the subcellular localisation and the protein level of overexpressed proteins. Transgene expression was induced during larval development with the en-Gal4 (engrailed) driver and their resulting expression was assayed in imaginal wing discs. The epitope-specific antibody detected the protein in the posterior compartment without obvious fluctuations in total abundance (Fig. 2A). Furthermore, protein abundance in the uninduced parts of the disc was below the detection limit of our staining protocol. It was also possible to identify subcellular expression patterns: nuclear (Rbf) and cytoplasmic membrane localisation (Fz) could be distinguished (Fig. 2A).

We also compared protein abundance in adult tissue samples. We chose components of the multiprotein COP9 signalosome (Wei et al., 2008) and tested their individual abundance by western blot assays after misexpression in eye tissue with GMR-Gal4. The proteins are specifically detected without significant signs of degradation products. The total abundance, however, varies between the different proteins, probably owing to differences in translation rates and protein turnover (Fig. 2B).

In summary, these functional tests demonstrate that overexpression of HA epitope-tagged proteins typically, but not always, results in phenotypes similar to the untagged clones. The results also point out the need to evaluate potential effects of epitope tags with additional experiments. A detailed study validating the biological usefulness of this UAS-ORF library is presented elsewhere (Schertel et al., 2013).

Cloning sequence-verified ORFs at a genome-wide scale into a system suitable for in vivo use represents a challenge for creating an ORFeome library. Importantly, the ORFs should be in a system that guarantees the flexible re-use of the ORFs without the need to re-amplify and re-verify them.

The Berkeley Drosophila Genome Project (BDGP) offers such a resource (Yu et al., 2011). The BDGP has generated thousands of sequence-verified ‘movable ORFs’, either with a native stop codon or without such a stop for C-terminal tagging. These ORFs are provided in a Cre/loxP-based gene transfer system, also known as Creator System (Clontech). We tested this system to make use of this ORF collection, but found it unsatisfactory in our in vivo misexpression tests. In our hands, the splicing required to attach the tag to the ORF was unspecific, generating multiple splice variants. Compared with our 3xHA-tagged Gateway clones, this resulted in lower expression of the correct splice version and therefore explains the complete absence of misexpression phenotypes (a detailed assessment of this approach can be found in supplementary material Appendix S1 and Fig. S4).

For further development of the cloning system, we reverted to the Gateway technology to preserve the advantage of re-using ORFs for multiple purposes once they are cloned and sequence verified. However, we made some modifications to the 3xHA-tagged destination vector by introducing two shortened and mutated FRTs (FLP recognition targets, namely FRT2 and FRT5) (Schlake and Bode, 1994), which immediately flank the Gateway cassette (Fig. 3A). Recombination between these mutant FRT sites is reported to be incompatible (Larsen et al., 2006), which is essential for preventing the excision of the intervening sequence in presence of a FLP recombinase (FLP).

Because incompatibility between the selected FRT sites is crucial for our system, we performed multiple tests of the behaviour of the two FRT variants with clonal GFP assays in vivo (supplementary material Fig. S3). We examined each of the FRTs for recombination with an identical FRT, between each other, and together with a wild-type FRT. These tests were carried out in an intra-chromosomal fashion (supplementary material Fig. S3), essentially to exclude any recombination between the two mutated FRTs, and in an inter-chromosomal fashion with the purpose of demonstrating that efficient recombination does occur in trans between identical FRTs (not shown). We confirmed that the two FRT variants were incompatible, supporting the use of the selected FRTs for our promoter and tag swapping strategy, described below.

The two mutant FRTs allow distinct modification of the UAS-ORF library by FLP/FRT recombination in vivo. The C-terminal 3xHA tag can be exchanged with any other tag of choice, e.g. to switch to another epitope/antibody pair. Similarly, the UAS promoter region can be exchanged with any other promoter, including the possibility of adding N-terminal tags. The swapping events can be tracked by specific gain and loss of markers, e.g. the C-terminal exchange events can be recovered by selecting for w⁺y⁺ recombinants, the N-terminal exchange by selecting for y⁺ (see Fig. 3B). Furthermore, N- and C-terminal exchanges can be combined sequentially by adjusting the markers in the ‘promoter’ and ‘3′-tagging’ lines accordingly. For example, a combined promoter swap and tag swap could be achieved by first initiating the promoter swap and selecting for y⁺ and then performing a tag swap with a w⁺ marked tag-line and choosing y⁺w⁺ recombinants. The swapping device makes the ORF lines highly flexible for different applications and assays. Repeated injections to obtain new constructs are avoided after the one-time creation of the ORF and the specific swapping lines.

Currently, we have created two promoter lines (for lexO and actin5C) and three C-terminal tag lines (for eGFP and the two Venus YFP fragments VNm9 and VC155) at cytological position 86F that are compatible with our library. All five lines carry an hsp70-FLP construct on the X chromosome, necessary for the FLP/FRT inter-chromosomal recombination.

To test the efficiency and accuracy of this system, we generated a few UAS-ORF lines with a vector containing all the features depicted in Fig. 3A, named pTF-HA.attB and later upgraded to pGW-HA.attB. One of these lines, containing bicoid (bcd), was used to test swapping rates: males of UAS-bcd were crossed to either lexA operator (lexO) or to eGFP tag females, followed by a single 30-minute heat shock of the progeny at day 3; hatching females were outcrossed (four females per vial; F1 males can be used similarly, see Materials and methods) and the progeny was scored for the appropriate marker combinations, indicating successful swapping events. Under these conditions, the lexO promoter swapping resulted in recombinants found in 84% of the F1 crosses (n=39), whereas the exchange of the C-terminal tags occurred in 71% of the F1 crosses (n=24). These high rates are achievable with a convenient vial setup and a single heat-shock treatment. Finally, we confirmed the accuracy of the exchanges by PCR amplification and sequencing of the involved regions undergoing recombination (not shown).

Next, we validated the swapping by measuring protein expression before and after the swap (Fig. 4). Misexpression of UAS-bcd^HA by dpp-Gal4 resulted in a distinct expression domain along the anterior/posterior (A/P) axis (Fig. 4A). Following a UAS-hsp70 for lexO-hsp70 promoter exchange, we overexpressed the bcd^HA transgene with the lexA TA (transactivator) driver dpp-LHG (Yagi et al., 2010), leading again to the distinct expression along the A/P axis (Fig. 4B). In this LHG-driven misexpression experiment, the wing discs displayed a pronounced fold in the wing pouch domain, different to the Gal4-driven expression for which we observed only a weak morphological effect. Misexpression of bcd^HA with MS1096-Gal4 is lethal in males and causes significantly smaller wings in females (not shown). Thus, the observed morphological abnormality in the wing discs (Fig. 4B) is an early sign of the detrimental effects of bicoid misexpression. This early morphological alteration only appears weakly in the Gal4-driven bicoid expression (Fig. 4A), probably because the transcriptional activation by the Gal4 driver is substantially weaker than that observed for the driver LHG (Yagi et al., 2010). To visualise the 3′-tag exchange we used the transgene UAS-emc (extra macrochaetae) and monitored the expression of the protein via the 3xHA tag before (Fig. 4C) and after (Fig. 4D) the exchange with the eGFP tag. Both proteins, Emc^HA and Emc^eGFP, led to identical expression domains when driven by the en-Gal4 line.

The ability to test for direct physical interactions between selected proteins with bimolecular fluorescence complementation (BiFC) analysis further demonstrates the versatility of the FRT-mediated swapping tool (Hu et al., 2002; Saka et al., 2007). This method visualises direct protein interactions by reconstituting a functional YFP from two non-fluorescent subfragments of YFP, which are fused to interacting proteins. We tested a previously documented (Jaw et al., 2000) (Fig. 4E) and a potential (Fig. 4F) protein-protein interaction by C-terminally tagging one ORF with the Venus YFP fragment VNm9 and the other with fragment VC155, followed by co-expression of the fusion proteins in the eye tissue by GMR-Gal4. A strong fluorescence signal was detected upon co-expression of HTH-VNm9 and EXD-VC155 (i.e. Homothorax and Extradenticle; Fig. 4E), confirming in vivo a direct protein-protein interaction as previously documented (Jaw et al., 2000). No specific fluorescence was detected when testing for interaction between HLHm5-VNm9 and HLH54F-VC155 (Fig. 4F).

Taking together, the modified FRT sites allow us to manipulate the promoters and tags of the library in vivo and validation tests have shown that this works accurately and efficiently. The BiFC analysis is one demonstration of using the swapping device, in this case to gain knowledge of potential in vivo protein-protein interactions.

Here, we provide an efficient strategy for creating a UAS-ORFeome library in Drosophila. Our effort to create such a library is mainly motivated by the experience that the majority of genes do not show ‘obvious’ loss-of-function phenotypes in conventional genetic screens (e.g. Miklos and Rubin, 1996). We propose that comprehensive and efficient gain-of-function screens using a UAS-ORF library will be a useful approach towards uncovering phenotypes for genes that have eluded loss-of-function screens. We extensively tested our approach with a pilot library of ~1200 transgenic lines and optimised the cloning, embryo injections and in vivo expression. The high integration rates and site specificity achieved with the ΦC31 integrase system reduces production time and improves the overall quality of the transgenic library, while avoiding the labour-intensive mapping required with traditional approaches. Molecular barcodes and plasmid pool injections further streamline the whole process.

The strategy presented here is a powerful way to create a genome-wide library for gain-of-function screens. However, some aspects require further consideration. The comparison between native and 3xHA-tagged clones highlights the problem of using tagged lines. Although the vast majority of the C-terminally 3xHA-tagged proteins behave like the untagged counterparts, we might miss up to 20% of candidates in a screen because of this epitope tag (false negative). In cases in which the tagged version causes a phenotype contrary to the wild-type protein (false positive), the tag might alter the protein stability, e.g. by masking a domain that is required for degradation. Potential interference of tags with protein function, for example leading to misrouting, has been demonstrated (Romano et al., 1998; Brothers et al., 2003). Furthermore, the accurate subcellular localisation of a protein might also be altered simply by overexpression. Approaches that tag proteins in their endogenous loci are likely to be more reliable strategies to reflect the accurate subcellular localisation (see Venken et al., 2011). Together, these results and considerations strongly emphasise the importance of including appropriate controls to address the effects of epitope tags. Despite this limitation, we favour the creation of a tagged version of the library, as this facilitates further analysis, such as immunohistochemistry, mass spectrometry or ChIP assays.

Generally, dominant-negative effects can result from overexpression (Herskowitz, 1987). For example, components of multiprotein complexes might be more prone to this effect, as an imbalance in the subunit composition can be fatal. Suspected dominant-negative effects, however, could be confirmed by screening corresponding mutants or RNAi lines.

Despite these limitations, this growing UAS-ORFeome library greatly extends the currently available tools for elucidating gene function in Drosophila. The Gal4/UAS system, clearly the most widely used binary expression system in Drosophila genetics, provides a rich repertoire of cell- and tissue-specific expression, which is continuously being expanded. Overexpression levels can be tuned in various ways and, therefore, using the UAS-ORF library will allow a broad range of phenotypic read-outs. Furthermore, this resource can also be used in combination with other genetic tools, such as the MARCM (mosaic analysis with a repressible cell marker) technique (Lee and Luo, 1999). Doing overexpression screens in modified backgrounds will often be the reasonable strategy. Unlike loss-of-function phenotypes, interpreting overexpression phenotypes will usually be more challenging (Prelich, 2012). For example, whether an observed phenotype results from an activating or inhibiting mechanism has to be determined. Testing the candidates for loss-of-function effects (i.e. mutants, RNAi) will be a likely next step.

The introduced FLP/FRT-mediated swapping technology is a convenient method for customising the library towards specific applications, as the appropriate tagging constructs need to be injected only once. Owing to the high exchange efficiencies, many transgenic lines can be converted in parallel. A very basic application is the creation of fluorescence-tagged transgenes for expression and localisation studies. This swapping device is also optimally suited to test potential protein-protein interactions in vivo with the BiFC method, as we have demonstrated. Recently, different protein inactivation methods were developed, such as ‘TEV protease-induced protein inactivation’ (TIPI) (Taxis et al., 2009) and ‘degrade Green Fluorescent Protein’ (deGradFP) (Caussinus et al., 2011). These methods require the attachment of a degron unit to the target protein or the creation of a GFP (or close derivative) fusion with the target protein, respectively. With the swapping device implemented in the library, this can be easily achieved. Generally, any sequence-encoded N- or C-terminal modification can be swapped onto a target protein with this exchange system.

The full genome-wide ORFeome library will take many years to complete. Here, we present the first set of UAS-ORF lines together with the detailed description of the technology underlying this project. The current subset comprises lines carrying cell cycle and growth control genes and, thus, is an interesting stand-alone library that can be readily used for various overexpression screens. These UAS-ORF lines will become publicly available (http://www.flyorf.ch) and new sub-collections, such as kinases/phosphatases and transcription factors, will be added continuously. In the longer term, we may consider expanding the scope of the library to also include isoforms, specifically altered sets of genes (e.g. constitutively active or catalytically inactive mutant forms) or heterologous gene sets. This ORFeome project offers a powerful, new and continuously expanding resource for systematically uncovering and testing the function of genes in any given genetic pathway.

We thank Kenneth H. Wan, Mark Stapleton and Susan E. Celniker for a pDNR-Dual vector and information; Sini Miettinen, Aileen McLeod, Reto Städeli, Eliane Escher, Nellcia Wang and Cristina Bastos for technical assistance; Robert P. Zinzen, David Luginbuhl and Liqun Luo for expression analysis of our UAS-lacZ-attP lines; Jean-Paul Vincent for information on the mutated FRT versions; Damian Brunner for alerting us to an alternative linker sequence; Roland Baumgartner for advice and reagents for BiFC; Korneel Hens and Bart Deplancke for providing information on specific protein-protein interactions. We are very grateful to Jeremiah J. Zartman and Ludovic Baillon for critical reading of the manuscript.