Expression of varied GFPs in Saccharomyces cerevisiae: codon optimization yields stronger than expected expression and fluorescence intensity

Green fluorescent protein (GFP), which was originally isolated from jellyfish, is a widely used tool in biological research, and homologs from other organisms are available. However, researchers must determine which GFP is the most suitable for a specific host. Here, we expressed GFPs from several sources in codon-optimized and non-codon-optimized forms in the yeast Saccharomyces cerevisiae, which represents an ideal eukaryotic model. Surprisingly, codon-optimized mWasabi and mNeonGreen, which are typically the brightest GFPs, emitted less green fluorescence than did the other five codon-optimized GFPs tested in S. cerevisiae. Further, commercially available GFPs that have been optimized for mammalian codon usage (e.g., EGFP, AcGFP1 and TagGFP2) unexpectedly exhibited extremely low expression levels in S. cerevisiae. In contrast, codon-optimization of the GFPs for S. cerevisiae markedly increased their expression levels, and the fluorescence intensity of the cells increased by a maximum of 101-fold. Among the tested GFPs, the codon-optimized monomeric mUkG1 from soft coral showed the highest levels of both expression and fluorescence. Finally, the expression of this protein as a fusion-tagged protein successfully improved the reporting system’s ability to sense signal transduction and protein–protein interactions in S. cerevisiae and increased the detection rates of target cells using flow cytometry.

coefficient (ε ) and two excitation maxima (395 nm and 475 nm) [16][17][18] . Enhanced GFP (EGFP), a mutant of the original GFP that was developed through mutagenesis studies, is among the most popular and widely used monomeric GFPs, and it exhibits a single excitation peak at 488 nm 19 (Table 1). EGFP contains codons that were optimized for mammalian cell expression and S65T and F64L mutations that improve the spectral characteristics, fluorescence intensity and stability of the protein and increased protein maturation efficiency at 37 °C 19,20 .
Recently, GFP variants have been discovered in a wide range of sources, including various Aequorea species 21,22 , copepods 23 , amphioxus 24,25 and reef corals [26][27][28][29] , and these variants are now commercially available. For example, AcGFP1, a brilliant green monomeric GFP variant isolated from Aequorea coerulescens, exhibits 94% sequence homology to EGFP 21 (Table 1). The monomeric and pH-stable GFP, TagGFP2 (mTagGFP), which is the improved variant of TagGFP derived from Aequorea macrodactyla, exhibits 80% sequence homology to EGFP 22 (Table 1). The coral-derived GFPs mUkG1 26 , ZsGreen 27 and mWasabi 29 , bear surprisingly low sequence similarity to the GFPs that were originally identified in Aequorea species. Monomeric mUkG1 emits intense fluorescence and is extremely stable against high pH, owing to the introduction of five mutations to the dimeric wild-type fluorescent protein UkG1, which was isolated from the soft coral Sarcophyton sp. 26 (Table 1). Tetrameric ZsGreen is a Zoanthus sp. fluorescent protein variant with a single amino acid substitution that enhances the emission characteristics; the protein has a substantially higher quantum yield (QY) 27 (Table 1). The bright, monomeric green fluorescent protein mWasabi is a variant of monomeric mTFP1, which is derived from the tetrameric cyan fluorescent protein cFP484 of Clavularia sp. 29 (Table 1). mNeonGreen is a monomeric yellow-green fluorescent protein that emits the brightest green or yellow fluorescence reported to date; this protein was derived from the tetrameric yellow fluorescent protein (LanYFP) of the cephalochordate Branchiostoma lanceolatum by structure-guided directed evolution 24 (Table 1).
Thus, a wide range of GFPs are now available. These proteins generating green fluorescence are crucially important as versatile experimental tools, because of their compatibilities for varied fluorometric instruments. Codon optimization has reported to increase the expression levels of several fluorescent proteins in S. cerevisiae [30][31][32] . However, the subjects of investigation were limited to the EGFP or other major color fluorescent proteins, and the choice of a suitable GFP for expression in S. cerevisiae is less well studied among a wide variety of GFPs.
Here, we expressed seven codon-optimized GFPs and five non-codon-optimized GFPs in S. cerevisiae and compared the protein expression levels and green fluorescence intensities. Through the evaluations, we determined the nucleic acid sequences of the codon-optimized GFPs that were expressed at high levels and that emitted intense green fluorescence in S. cerevisiae. Furthermore, we tested these GFPs that were expressed at relatively high levels for use as reporter genes in fusion form for detecting signal transduction and protein-protein interactions. Finally, we demonstrated the codon-optimized mUkG1 fusion reporter successfully increased the detection rates of target cells using flow cytometry.

Results and Discussion
Expression of GFPs in S. cerevisiae. We compared the expression of seven GFPs from various sources in S. cerevisiae: EGFP, AcGFP1, TagGFP2, mUkG1, ZsGreen, mWasabi and mNeonGreen ( Table 1). The genes encoding the seven GFPs were optimized for expression in the yeast S. cerevisiae (yEGFP, yAcGFP1, yTagGFP2, ymUkG1, yZsGreen, ymWasabi and ymNeonGreen) (Supplementary Table S1) and inserted into the multiple-cloning site of a pGK416 single-copy autonomous replicating plasmid for constitutive expression under the control of the PGK1 promoter 33 (Fig. 1A and Table 2). The yeast BY4741 strain was transformed with the constructed plasmids (Tables 2 and 3 and Supplementary Table S2), and their green fluorescence was evaluated using flow cytometry and fluorescence microscopy ( Fig. 1 and Supplementary Table S3).
The green fluorescence intensities of 10,000 cells were quantitatively measured using flow cytometry, and the values were recorded as the mean fluorescence intensities (MFIs). Monomeric ymUkG1 and tetrameric yZsGreen exhibited the highest fluorescence intensities over 40,000-45,000 MFIs (Fig. 1B and Supplementary Table S3). In contrast, ymWasabi and ymNeonGreen exhibited remarkably low fluorescence intensities (less than 20,000 MFIs) (Fig. 1B and Supplementary Table S3), even though both GFPs are the brightest green fluorescent proteins (mWasabi, ε = 70,000 M −1 cm −1 , QY = 0.80; mNeonGreen, ε = 116,000 M −1 cm −1 , QY = 0.80) 24,29 (Table 1). Fluorescence microscopy observations also showed that the green fluorescence of these two proteins was less  Table 1. Properties of green fluorescent proteins tested. Along with the common name for each GFP, the peak excitation (λ ex) and emission (λ em) wavelengths, molar extinction coefficient (EC), quantum yield (QY), relative brightness (RB), association state and reference or source are listed. Brightness values were derived from the product of EC and QY, divided by the value for EGFP.
intense than that of the other five GFPs (Fig. 1C). Therefore, we decided to use GFPs other than mWasabi and mNeonGreen (EGFP, AcGFP1, TagGFP2, mUkG1 and ZsGreen) in the later experiments. Western blot analysis was performed using as the primary antibody monoclonal anti-frag M2 antibody for the GFPs fused frag tag. Alkaline phosphatase-conjugated anti-mouse IgG was used as the secondary antibody, and colorimetric detection of alkaline phosphatase activity was performed using CDP-Star detection reagent. 'Control' indicates the BY4741 yeast strain harboring a mock pGK416 plasmid.   Next, we compared the expression of the codon-optimized sequences and the commercially available (non-optimized for S. cerevisiae) sequences of five GFPs (Table 2 and Supplementary Table S2) by flow cytometry, fluorescence microscopy and western blot analysis ( Fig. 1B-D). Of the commercially available GFP sequences tested, EGFP, AcGFP1 and TagGFP2 include codons that were optimized for mammalian cell expression, and ZsGreen and mUkG1 used the codons from the isolated, wild type organism. The flow cytometry and fluorescence microscopy observations for codon-optimized GFPs are shown in Fig. 1B Table S3 and Fig. S1). Fluorescence microscopy revealed clear differences in the green fluorescence intensity of the codon-optimized and non-codon-optimized GFPs (Fig. 1C). Using the FLAG tag fused GFPs (Table 2 and Supplementary Table S2), the western blot analysis showed increased expression levels of the codon-optimized GFPs (approx. 27 kDa) in S. cerevisiae (Fig. 1D). Moreover, intriguingly, all yeast cells expressing codon-optimized GFPs generated light green colors under natural light, unlike non-codon-optimized EGFP, AcGFP1 and TagGFP2, which were never green under natural light ( Supplementary Fig. S2).
As shown in Table 4, codon-optimized GFPs for S. cerevisiae presented high codon adaptation indexes (CAIs; 0.87~0.90) and moderate GC contents (39.7~40.4%). In contrast, commercially available, non-codon-optimized EGFP, AcGFP1 and TagGFP2 sequences unexpectedly exhibited much lower CAIs (0.50~0.56) and higher GC contents (59.0~63.6%) ( Table 4), although the codon usage had been optimized for mammalian expression. This may have misled us into thinking that the mammalian cells had reasonably similar codon usages to the eukaryotic organism S. cerevisiae. For mUkG1 and ZsGreen, the wild type codons from the soft coral Sarcophyton sp. and the button polyp coral Zoanthus sp. were used, unexpectedly resulting in a relatively similar codon usage to that in yeast (CAIs = 0.69~0.70 and GC contents = 44.9~45.0%) ( Table 4). These support our previous result that ZsGreen (original codons) exhibited more intense fluorescence than EGFP (non-codon-optimized) when used as a fluorescent reporter protein 34 . Among the codon-optimized GFPs tested in this study, monomeric ymUkG1 and tetrameric yZsGreen exhibited the most intense green fluorescence in S. cerevisiae (Fig. 1B).
GFP expression when used as fusion-tagged proteins to report the activation of signal transduction in S. cerevisiae. GFP is most widely applied as a protein tag or as a reporter gene 4,7,32,35,36 . To determine the applicability of the GFPs tested above, we decided to test the pheromone-responsive FIG1 gene, which is involved in conjugation and cellular fusion on plasma membrane during the mating process 37 . FIG1 gene expression is induced in response to activation of the yeast mating signaling pathway; therefore, this gene has been often used to sense the signal promoted in the mating pathway ( Fig. 2A) or the agonist response to heterologous GPCRs using the non-codon-optimized EGFP as the reporter gene [38][39][40][41] .
The five codon-optimized and non-codon-optimized GFP genes were integrated into the chromosome of BY4741 haploid yeast a-cells to fuse to the C-terminus of the genomic FIG1 gene (Table 3 and Supplementary  Table S4). To induce the transcription of FIG1-GFPs, the cells were incubated for 6 hours in YPD medium containing the a-cell-specific mating pheromone (α -factor) ( Fig. 2A). The cultured cells were then analyzed by flow cytometry and observed under a fluorescence microscope (Fig. 2B,C).
The GFPs tested as fusion reporters exhibited marked increases in green fluorescence in response to the addition of α -factor pheromone (Fig. 2B). Although these fusion proteins showed lower fluorescence intensities than those of the solely expressed untagged GFPs, the MFIs of the fusion-codon-optimized GFPs were greater than those of all five non-codon-optimized GFPs (Fig. 2B). The pheromone-stimulated cells induced 22.8~110.0-fold changes in the MFIs compared to the corresponding unstimulated cells (Fig. 2B). The MFIs of Fig1-GFPs were increased approximately 1.6~2.0-fold due to codon optimization. Codon-optimized ymUkG1 and yZsGreen exhibited higher fluorescence intensities than the most commonly used non-codon-optimized EGFP when expressed as fusion genes with the FIG1 reporter gene (FIG1-EGFP, MFI = 1,469; FIG1-ymUkG1, MFI = 6,184; FIG1-yZsGreen, MFI = 7,143) and exhibited greater than 4.2-and 4.9-fold increases (Fig. 2B). Fluorescence microscope observation revealed brighter green fluorescence in pheromone-stimulated yeast cells encoding Fig1-ymUkG1 and Fig1-yZsGreen than in cells encoding Fig1-EGFP (the cell shapes were elongated by the excessive pheromone stimulation) (Fig. 2C). In addition, Fig1-ymUkG1 fusion protein (monomeric GFP fusion) especially showed correct localization on the plasma membrane, although it has been also observed in other organelle caused by the extensive overexpression (Fig. 2C).
As shown above, similarly to the results obtained for the solely expressed untagged GFPs, codon-optimized ymUkG1 and yZsGreen exhibited the brightest fluorescence, even when expressed as fusions with the FIG1 reporter gene. Tetrameric fluorescent proteins are generally regarded as unfavorable for use as fusion tags because they interfere with the normal function of the target proteins 4 ; however, tetrameric yZsGreen could function as the quantitative fusion reporter, although the fusion with the Fig1 protein rarely localized on the plasma membrane. In further experiments, we studied monomeric ymUkG1, which is apparently less constraining in many applications but exhibits significantly more intense fluorescence than other monomeric GFPs in S. cerevisiae.
Scientific RepoRts | 6:35932 | DOI: 10.1038/srep35932 Application of codon-optimized ymUkG1 in Gγ recruitment systems that report protein-protein interactions in S. cerevisiae. We developed the Gγ recruitment system to detect protein-protein interactions (PPIs) 40,[42][43][44][45] and to screen for protein variants presenting desirable affinities 40 . The detection method of the Gγ recruitment system is based on the fundamental principle that yeast pheromone (mating) signaling requires the localization of a complex between guanine nucleotide binding protein (G-protein) β -and γ -subunits (Gβ γ ) and the inner leaflet of the plasma membrane 46 . In brief, for the machinery to detect PPIs, an engineered Gγ mutant (termed Gγ cyto ) is used that lacks a membrane localization sequence (lipidation motif); this localization signal is normally expressed in the cytosol. When the target protein (X) is a soluble cytosolic protein, Gγ cyto is prepared as a fusion protein with the target protein (X) (Gγ cyto -X), and library proteins (Y) are attached to the artificial lipidation site to localize the proteins to the membrane ( Supplementary Fig. S3A) 40,42 . In contrast, when the target protein (X) is a membrane-bound protein, Gγ cyto is prepared as a fused protein with the library proteins (Y) (Gγ cyto -Y) ( Supplementary Fig. S3B) 43,45 . When target "X" and candidate "Y" interact, the Gγ cyto -X ( Supplementary Fig. S3A) or Gγ cyto -Y ( Supplementary Fig. S3B) fusion protein targets the Gβ to the membrane and subsequently induces pheromone-signaling pathway activation. Signaling can be monitored by a fluorescent reporter assay or a mating growth assay; therefore, the PPIs are easily detected (Supplementary Fig. S3C).
Here, FIG1-EGFP (non-optimized) was used as a fluorescent reporter gene in the Gγ recruitment system. If the flow cytometer (cell sorter) that can analyze more than several thousand cells per second is available, the fluorescent reporter assay extremely speeds up and is helpful for screening PPI protein variants. We tested    whether FIG1-ymUkG1 could improve the fluorescence reporting ability of the Gγ recruitment system. As for the previously described system, the Fc protein of human immunoglobulin G (IgG) and the Z domain (derived from Staphylococcus aureus protein A (Z WT )) 47 were used as the PPI models. Several Z variants (Z K35A , Z I31A and Z 955 ) with differing affinities for the Fc protein were also used as the PPI models (Z WT , 5.9 × 10 7 M −1 ; Z K35A , 4.6 × 10 6 M −1 ; Z I31A , 8.0 × 10 3 M −1 ; Z 955 , none) 48,49 . We first tested the use of soluble cytosolic target proteins ( Supplementary Fig. S3A). The Fc protein was used as the cytosolic target protein 'X' and was fused with Gγ cyto (Gγ cyto -Fc). Four Z domain variants were attached to the artificial lipidation motifs (derived from Ste18 p) as the candidate 'Y' proteins, which were localized on the membrane (Tables 2 and 3 and Supplementary Table S5). After the addition of α -factor pheromone, the cells were cultured, and the GFP signal was measured using flow cytometry (Supplementary Fig. S3C). Both reporters (Fig1-EGFP and Fig1-ymUkG1) successfully detected PPIs between Fc and several Z variants (Z WT , Z K35A and Z I31A ) (Fig. 3A). However, the fluorescence intensities of the two reporters differed greatly. When comparing the systems in Z WT , the Next, we tested the use of membrane target proteins ( Supplementary Fig. S3B). The Fc protein was used as the membrane target protein 'X' . To determine the positions at which to fuse the lipidation motifs, both C-terminal and N-terminal membrane-associated Fc proteins were prepared (a C-terminal lipid anchor, derived from Ste18 p; and a N-terminal lipid anchor, derived from Gpa1p). Four Z domain variants were used as cytosolic candidate 'Y' proteins and fused with Gγ cyto (Gγ cyto -Z variants) ( Table 3 and Supplementary Table S6). The cells were assayed similarly to the previous experiments. For the C-terminal membrane-associated Fc proteins, both the Fig1-EGFP and Fig1-ymUkG1 reporters successfully detected the PPIs with two Z variants (Z WT and Z K35A ), although they did not detect the PPI with Z I31A (Fig. 3B). This result was consistent with previous results that failed to detect a PPI between membrane-associated Fc and Gγ cyto -fused Z I31A . The weak PPI (8.0 × 10 3 M −1 ) is likely below the detection limitation in the Gγ recruitment system when using Fc as the membrane target protein 45 . Similarly to the results of the test for the soluble cytosolic target Fc protein (Fig. 3A), a large difference was observed in the fluorescence intensities between the Fig1-EGFP and Fig1-ymUkG1 reporters (Fig. 3B). When comparing the variants in Z WT , the Fig1-EGFP and Fig1-ymUkG1 reporters emitted fluorescent signals that were 40-and 131-times more intense than those of the respective negative control cells. The fluorescence intensities were 3.6-fold different between Fig1-EGFP (MFI = 2,380) and Fig1-ymUkG1 (MFI = 8,519). When comparing the variants in Z K35A , the difference in fluorescence intensity between the Fig1-EGFP and Fig1-ymUkG1 reporters was 2.4-fold. When using the N-terminal membrane-associated Fc proteins, similar tendencies were observed ( Supplementary Fig. S4). When using Z WT and Z K35A, the differences in fluorescence intensity between the Fig1-EGFP and Fig1-ymUkG1 reporters were 3.6-and 2.9-fold, respectively. Thus, when used as a fluorescence reporter, codon-optimized ymUkG1 yielded brighter fluorescence than the previous non-codon-optimized (for S. cerevisiae) EGFP, even in the Gγ recruitment systems, which detect PPIs for both soluble cytosolic target proteins and membrane target proteins.

FACS analysis of target cells in cell mixtures containing positive and negative cells.
Fluorescence activated cell sorting (FACS) is a specialized instrument that enables analysis and sorting of rare target cells in a larger cell population. To evaluate the performance of ymUkG1 as the reporter for FACS analysis, the Gγ recruitment system for soluble cytosolic target proteins was used ( Supplementary Fig. S3A and Fig. 3A).
To determine the gate area that dominantly included positive cells (GFP + ), we analyzed the data of yeast strains expressing Gγ cyto -Fc and membrane-anchored Z variants (Z WT , Z K35A , Z I31A and Z 955 ), whose fluorescence was acquired using a flow cytometer in Fig. 3A. We set the gate area (GFP + ) that completely excluded negative cells (expressing Z 955 with Gγ cyto -Fc) and determined the positive cells by making the dot plots (forward scatter,  Fig. S5). The proportions of positive cells included in the 10,000 cells (expressing Z WT , Z K35A and Z I31A with Gγ cyto -Fc) using the Fig1-ymUkG1 reporter were respectively 98.3%, 95.4% and 80.3%, while those using the Fig1-EGFP reporter were 58.8%, 13.9% and 2.5% ( Supplementary  Fig. S5 and Table S7). This indicates the Fig1-ymUkG1 reporter clearly displayed higher percentages of the cells that were defined as positive cells (GFP + ) than the Fig1-EGFP reporter, even though both cells expressing the same interaction pairs could be presumed to coincide in the promoted signaling and GFP transcription levels.

FSC-A vs green fluorescence, GFP-A) (Supplementary
To further demonstrate the ability of the ymUkG1 reporter, we prepared the cell mixtures composed of positive cells (expressing Z WT with Gγ cyto -Fc) and negative cells (expressing Z 955 with Gγ cyto -Fc) with two different mixing rates (0.1% and 0.01% positive cells). After cultivation of the cell mixtures in the presence of α -factor for 6 hours, 10 6 cells were analyzed using flow cytometer and the content rates of positive (target) cells contained in the gate area (GFP + ) were evaluated (Fig. 4 and Table 5). Both Fig1-EGFP and Fig1-ymUkG1 reporters exposed the target cells contained in the GFP + area, whereas they showed the obvious distinctions in the detection rates of the target cells ( Fig. 4 and Table 5). When using the cell mixtures expressing Z WT and Z 955 with Gγ cyto -Fc (initially contained 0.1% and 0.01% positive cells), Fig1-ymUkG1 reporter detected 58% and 56% of the target cells (percentages of detected target cell numbers in GFP + area against initial cell numbers of positive cells), while Fig1-EGFP reporter detected 33% and 34% of the target cells, respectively (Fig. 4A-D and Table 5). When using the cell mixtures expressing Z I31A and Z 955 with Gγ cyto -Fc (in the case of weak interaction), more clear differences were found between Fig1-EGFP and Fig1-ymUkG1 reporters (Fig. 4E-H and Table 5). Briefly, Fig1-ymUkG1 reporter detected 74% and 65% of the target cells, while Fig1-EGFP reporter detected 12% and 24% of the target cells (for the cell mixtures containing 0.1% and 0.01% positive cells, respectively) (Fig. 4E-H and Table 5). The reason why the use of Z I31A displayed higher percentages of the detectable target cells than that of Z WT might have arisen from the growth biases during the cultivation of cell mixtures. It is well known that the pheromone signaling (in response to the PPIs) provokes the cell-cycle arrest in G1 phase to make arrangements for the mating 50 . Because the percentages of the detectable positive cells to the initially contained positive cell numbers directly affect the recovery rates of the hopeful target cells, they are crucially important factors on FACS sorting. Thus, we have successfully demonstrated the superiority of the codon-optimized ymUkG1 reporter compared to the non-codon-optimized EGFP reporter in S. cerevisiae.
In summary, we compared the expression of seven codon-optimized GFPs and five non-codon-optimized GFPs in S. cerevisiae. Codon optimization for S. cerevisiae improved all five non-codon-optimized GFPs and surprisingly increased the fluorescence intensities up to 101-fold compared with commercially available GFPs (which were optimized for mammalian codon usage). CAI and GC content affected the expression of GFPs in S. cerevisiae more than expected. Among the codon-optimized GFPs tested, monomeric ymUkG1 and tetrameric yZsGreen exhibited the most intense fluorescence in S. cerevisiae, and green color was observed visually under natural light conditions in cells expressing these proteins. The expression of monomeric ymUkG1 protein was higher than that of tetrameric yZsGreen protein. Both ymUkG1 and yZsGreen were prepared as fusion reporters tagged to the Fig1 protein for use in monitoring signal transduction, and these fluorescent reporters exhibited significantly greater utility than the previously used, non-codon-optimized EGFPs (which were optimized for mammalian systems). Finally, the monomeric ymUkG1 was successfully used to improve the PPI detection system using the Fig1 reporter and increased the percentages of the detectable target cells using flow cytometry.
The most commonly used GFP in S. cerevisiae is probably the commercially available EGFP, whose codons are optimized for mammalian cells, not S. cerevisiae. Because gene and fragment synthesis services are readily available, it is easy to prepare a DNA fragment containing codon-optimized nucleic acid sequences. When strong protein expression is required, it might be best to measure the CAI and GC content and to test the codon-optimization of the protein of interest. We propose that the codon-optimized monomeric protein ymUkG1 (Supplementary Table S1) is a good choice for GFP experiments involving S. cerevisiae.  Table 3. The yeast strains were grown in YPD medium containing 1% (w/v) yeast extract, 2% peptone and 2% glucose or in SD media containing 0.67% yeast nitrogen base without amino acids (BD-Diagnostic Systems, Sparks, MD, USA) and containing 2% glucose. Amino acids and nucleotides (20 mg/L histidine, 60 mg/L leucine, 20 mg/L methionine, or 20 mg/L uracil) were supplemented into SD media lacking the relevant auxotrophic components. Agar (2%; w⁄v) was added to the medium described above to produce YPD and SD solid media.
The plasmids used to integrate the GFP reporter genes at the FIG1 locus on the yeast chromosome were constructed as follows: A DNA fragment containing the homologous sequence of the FIG1 terminator (downstream of FIG1 gene; 200 bp) was PCR-amplified from pBlue-FIG1pt-ZsGreen 34 using the primer pair 35 and 36. The amplified fragments were digested with XhoI + KpnI and inserted into the pBlueScript II KS(+ ) Figure 3. Expression of non-codon-optimized EGFP and codon-optimized ymUkG1 as fusion-tagged proteins to report PPIs using Gγ recruitment systems. (A) Flow cytometry analyses using the Gγ recruitment system for cytosolic target proteins. The Fc protein was used as the cytosolic target protein 'X' and was expressed as a fusion protein with Gγ cyto (Gγ cyto -Fc). Membrane-anchored Z variants (Z WT , Z K35A, , Z I31A and Z 955 ) were expressed as 'Y' library candidate proteins. 'Control' indicates BFG2118 and UGFG2 yeast strains harboring the pGK413 mock plasmid (without the expression of 'Y'). (B) Flow cytometry analyses using the Gγ recruitment system for membrane protein targets. The Fc protein was used as the membrane target protein 'X' and was expressed as a membrane-associated protein with the C-terminal lipid anchor (derived from Ste18p). Four Z variants (Z WT , Z K35A, , Z I31A and Z 955 ) were used as cytosolic candidate 'Y' proteins and were expressed as fusion proteins with Gγ cyto (Gγ cyto -Z variants). 'Control' indicates the FC-G0 and UG2-FCG0 yeast strains (without the expression of 'Gγ cyto -Y'). The engineered strains were grown in media containing 5 μ M α -factor, and mean fluorescence intensities (MFIs) were analyzed. The MFIs of 10,000 cells were measured by flow cytometry.
The STE18 gene was substituted by kanMX4 in the yeast chromosome by amplifying the DNA fragment containing P STE18 -kanMX4-T STE18 (P STE18 : STE18 promoter and T STE18 : STE18 terminator) from pGK426-GPTK 42 using the primer pair 45 and 46. BYFUG2 was then transformed with the amplified DNA fragment, and transformants were selected on YPD solid medium containing G418 (500 ng/mL) (Nacalai Tesque, Kyoto, Japan) to yield the UGW2 strain.
DNA cassettes for the Gγ cyto -Fc protein in the cytosol were integrated as follows. DNA fragments containing URA3-P PGK1 -Gγ cyto -Fc-T PGK1 -T HIS3 (T HIS3 : HIS3 terminator) were amplified from pUMGP-Gγ MFcH 42 using the primer pair 61 (containing the homologous regions of the HIS3 promoter) and 62. UGW2 was transformed with the amplified DNA fragments using the lithium acetate method. The transformants were selected on SD-Ura plates (containing leucine, histidine and methionine) to yield UGFG2.
The constructed strains were used for the assays or plasmids were introduced using the lithium acetate method. All strains and transformants used for the assays are listed in Supplementary Tables S2-S5.

Culture conditions for GFP expression in yeast cells.
To express the tested GFPs (Fig. 1A), yeast transformants (Supplementary Table S2) were grown in 5 mL of SD-Ura medium at 30 °C overnight. The cultured cells were then inoculated into 2 mL of fresh SD-Ura medium to obtain an initial optical density at 600 nm (OD 600 ) of 0.1. The cells were thereafter cultivated at 30 °C for 3 hours and then harvested.
Signal transduction assays using FIG1-GFP reporter genes ( Fig. 2A) were conducted according to previously described methods 40 with some modifications. The engineered yeast a-cells (Supplementary Table S3) were grown in 5 mL of YPD medium 30 °C overnight. The cultured cells were then inoculated into 2 mL of fresh YPD medium containing 5 μ M α -factor (Zymo Research, Orange, CA, USA) to obtain an initial OD 600 of 0.1. The cells were thereafter cultivated at 30 °C for 6 hours and then harvested.
Protein-protein interaction assays using FIG1-GFP reporter genes were conducted as described previously 40 . In brief, for the Gγ recruitment system involving soluble cytosolic target protein (Fig. 3A), the engineered yeast a-cells (Supplementary Table S4) were grown in 5 mL of SD-His medium at 30 °C overnight. The cultured cells were then inoculated into 2 mL of fresh SD-His medium containing 5 μ M α -factor to obtain an initial OD 600 of 0.1. For the Gγ recruitment system involving membrane target proteins ( Fig. 3B and Supplementary Fig. S4), the engineered yeast a-cells (Supplementary Table S5) were grown in 5 mL of YPD medium at 30 °C overnight. The cultured cells were then inoculated in 2 mL of fresh YPD medium containing 5 μ M α -factor to obtain an initial OD 600 of 0.1. In both cases, the cells were thereafter cultivated at 30 °C for 6 hours and then harvested.
Fluorescence microscopy. Fluorescence microscopy observation was performed as described previously 53 .
The harvested cells were washed and resuspended in distilled water; the cell suspensions were then observed using a BIOREVO BZ-9000 fluorescence microscope equipped with a 60× objective lens (Keyence, Osaka, Japan). Flow cytometry. Flow cytometry analysis was conducted as described previously 54 . The harvested cells were diluted into test tubes containing the sheath solution, and GFP fluorescence was measured by using a BD FACSCanto II flow cytometer (BD Biosciences, San Jose, CA). The green fluorescence signal from 10,000 cells was excited with a 488-nm blue laser and collected through a 530/30-nm band-pass (GFP) filter. The data were analyzed by using BD FACSDiva software (version 5.0, BD Biosciences) and FlowJo software version 7.2.2 (Treestar, Inc., San Carlos, CA).  Table S4) were grown in 5 mL of SD-His medium at 30 °C overnight. Cell mixtures were prepared by mixing target yeast cells (BFG2118 + pGK413-ZWTmem, UGFG2 + pGK413-ZWTmem, BFG2118 + pGK413-ZI31Amem or UGFG2 + pGK413-ZI31Amem) and negative yeast cells (BFG2118 + pGK413-Z955mem or UGFG2 + pGK413-Z955mem) with 0.1% and 0.01% of the initial target cell ratios. The cell mixtures were then inoculated into 2 mL of fresh SD-His medium containing 5 μM α -factor to obtain an initial OD 600 of 0.1. The cells were thereafter cultivated at 30 °C for 6 hours, and then 10 6 cells were acquired by using a BD FACSCanto II flow cytometer and analyzed by using BD FACSDiva software.  Table 5. Summary of FACS analysis of cell mixtures containing positive and negative cells.