Arginine (R)-based ER localization signals are sorting motifs that confer transient ER localization to unassembled subunits of multimeric membrane proteins. The COPI vesicle coat binds R-based signals but the molecular details remain unknown. Here, we use reporter membrane proteins based on the proteolipid Pmp2 fused to GFP and allele swapping of COPI subunits to map the recognition site for R-based signals. We show that two highly conserved stretches—in the β- and δ-COPI subunits—are required to maintain Pmp2GFP reporters exposing R-based signals in the ER. Combining a deletion of 21 residues in δ-COP together with the mutation of three residues in β-COP gave rise to a COPI coat that had lost its ability to recognize R-based signals, whilst the recognition of C-terminal di-lysine signals remained unimpaired. A homology model of the COPI trunk domain illustrates the recognition of R-based signals by COPI.
The sorting of membrane proteins relies on different peptide-sorting motifs that are recognized by vesicle coats (like COPI, COPII, or clathrin) and their adaptor complexes. These interactions can capture cargo at various donor membranes (e.g., the ER, different Golgi compartments, or the plasma membrane), and lead to their inclusion into transport vesicles. Only for a few peptide-sorting motifs (e.g., motifs of the YXXΦ class that mediate the endocytosis of cargo proteins) is the interaction with the relevant clathrin adaptor complex understood in structural detail (Owen et al., 2004), whereas for many even the subunit harbouring the binding site remains unknown. ER localization signals comprising a C-terminal di-lysine motif (K(X)KXX) are thought to bind to the WD40 domains of the α- and β′-COPI subunits (Eugster et al., 2004). In contrast, it is completely unknown how other classes of ER localization motifs (e.g., R-based signals) are recognized. These signals occur in a growing list of cell surface receptors and ion channels (Michelsen et al., 2005). Many of these multimeric membrane proteins exert critical functions and therefore their assembly is tightly controlled to allow only multimers of a specific composition to reach the cell surface. Incompletely or incorrectly assembled complexes are retrieved back to the ER. In monomeric proteins R-based signals act as ER localization signals, but in assembled multimeric proteins they can be hidden in the complex or rendered inactive by the recruitment of 14-3-3 proteins (Zerangue et al., 1999, 2001; O'Kelly et al., 2002; Yuan et al., 2003; Brock et al., 2005; Heusser et al., 2006; Mrowiec and Schwappach, 2006). Previous work clearly implicates the COPI coat as an important player in the recognition of R-based signals: COPI was shown to bind to different types of R-based signals present in the two-pore domain potassium channel KCNK3 or to one subunit of the heterodimeric G protein–coupled receptor GABAB (O'Kelly et al., 2002; Brock et al., 2005). Yuan et al. (2003) showed specific (R-based signal-dependent) binding of purified mammalian coatomer to the tail of the inwardly rectifying potassium channel Kir6.2, the pore-forming subunit of ATP-sensitive potassium channels. Because coatomer forms a tight heptameric complex these experiments did not reveal the subunit containing the binding site for R-based signals. Having recently shown that R-based sorting motifs are functional in yeast (Michelsen et al., 2006), we embarked on a reverse genetic screen to characterize which genes are required for the function of R-based signals as ER localization signals. In addition, we dissected the role of COPI in the recognition of R-based signals. Our results demonstrate that combinatorial interaction with different binding sites may play an important role in the dynamic interplay between sorting machinery and membrane cargo proteins.
Results And Discussion
COPI-mediated retrieval is the main mechanism responsible for the activity of R-based signals as ER localization signals
We screened a deletion mutant library covering all nonessential Saccharomyces cerevisiae genes (Winzeler et al., 1999) and a yeast strain collection harboring essential genes under the control of a regulatable promoter (Mnaimneh et al., 2004). We fused the last 36 residues of the mammalian inwardly rectifying potassium channel Kir6.2 to Pmp2, a single-spanning (type I) yeast membrane protein, to obtain Pmp2GFP-LRKR. As shown previously (Michelsen et al., 2006), the reporter localized to the ER due to the exposure of the well-characterized R-based signal present in the Kir6.2 tail (Zerangue et al., 1999). We transformed a plasmid encoding the Pmp2GFP-LRKR reporter into both strain collections and analyzed the resulting transformants by light microscopy (Fig. S1 A). Inactivation of only 13 genes was found to affect the steady-state ER localization of the reporter (Table I). To further characterize the hits, a plasmid encoding Pmp2GFP-KKTN (Michelsen et al., 2006) was introduced into the strains identified in the screen (Fig. S1 A; and unpublished data). In wild-type cells, Pmp2GFP-KKTN localizes to the ER due to the exposure of the C-terminal di-lysine (KKXX) signal (Michelsen et al., 2006). R-based sorting motifs and KKXX signals fall into distinct classes because the details of how they are recognized are not identical (Shikano and Li, 2003; Michelsen et al., 2005). The ER localization of Pmp2GFP-LRKR and Pmp2GFP-KKTN was compromised in all candidate strains (Table I, Fig. S1 A; and unpublished data). Strikingly, several genes identified in the screen coded for COPI subunits or subunits of the Dsl1/Dsl3 tethering complex that is thought to be part of the ER target site for COPI-coated vesicles (Andag et al., 2001; Reilly et al., 2001). All but one coatomer subunit and the members of the Dsl1/Dsl3 tethering complex are essential. Hence, it is not surprising that the shut-off of these genes affects both types of cargo.
Because COPI functions as a heptameric complex, the genes encoding the α-, ε-, and ζ-COP subunit should have been identified in the screen. We investigated the localization of Pmp2GFP-LRKR and Pmp2GFP-KKTN in the corresponding deletion or shut-off strains (Fig. S1, B and C). The result shows that all COPI subunits but ε-COP (Fig. S2 A) are required for efficient ER localization of Pmp2GFP-LRKR. This result is consistent with the fact that ε-COP is the only dispensable COPI subunit (Duden et al., 1998). Presumably the genes encoding α- and ζ-COP were missed in the screen because of incomplete down-regulation. Similarly, the gene encoding Dsl1 that associates with Dsl3 to form a tethering complex thought to be involved in the fusion of retrograde vesicles with the ER was not detected in the screen, although the gene encoding Dsl3 was (Table I, Fig. S1 A). Consistent with the proposed role for a complex containing Dsl1 and Dsl3 in COPI-mediated retrograde transport, we found that down-regulation of the DSL1 gene resulted in the mislocalization of Pmp2GFP-LRKR and Pmp2GFP-KKTN (Fig. S1 B). The accumulation of coatomer subunits in the screen strongly suggests an intimate relationship between the sorting of membrane proteins exposing an R-based signal and coatomer. Therefore, we decided to concentrate on the role of COPI in the recognition of R-based signals.
Short conserved stretches in the δ- and β-COP subunits interact with R-based signals
The two WD40 domains of α- and β′-COP bind distinct but overlapping sets of di-lysine signals, and hence, both contribute to the recycling of proteins with di-lysine signals (Eugster et al., 2004). To assess whether coatomer discriminates between Pmp2GFP-LRKR and Pmp2GFP-KKTN, we tested the localization of both reporters in a mutant containing an allele with a mutation in the region of α-COP that encodes the WD40 domain (Fig. S2 B). Confirming previous results (Schroder et al., 1995; Eugster et al., 2004), we observed mislocalization of Pmp2GFP-KKTN, whereas Pmp2GFP-LRKR localized robustly to the ER. These experiments support the notion that the binding site for R-based signals present in the COPI complex is distinct from the WD40 domain of α-COP that recognizes the KKTN motif (Eugster et al., 2004).
The yeast two-hybrid system has been successfully used to detect the interaction between adaptor subunits and tyrosine-based cargo sorting motifs and between K(X)KXX motifs and COPI subunits (Ohno et al., 1995, 1998; Zerangue et al., 2001; Eugster et al., 2004). We used fusions to the Gal4 DNA-activation domain (AD) presenting a C-terminal tail with a very strong R-based signal (KLRRRRI) or one of low efficacy (NVRNRRK). Both synthetic signals were originally identified in a combinatorial screen of signal variants in mammalian cells (Zerangue et al., 2001). Their differential strength was revealed by quantitative trans-Golgi network processing and cell surface expression assays. Fusions of COPI subunits to the Gal4 DNA-binding domain (BD) (Faulstich et al., 1996) were used to assay the putative interaction between the R-based ER sorting motif and individual COPI subunits (Fig. 1 A). Two plasmids encoding the Gal4-BD fusions to β-COP or to δ-COP gave rise to specific activation of the HIS3 reporter gene (Fig. 1 A, “interaction test”) when tested against the strong variant of the R-based sorting motif.
We used the yeast two-hybrid assay to narrow down the sequence stretch on each COPI subunit that was required to observe a specific interaction with the strong R-based signal. A series of C-terminal deletion mutants of the Gal4-BD fusion to the δ-COP subunit was tested against the strong and the weaker R-based sorting motif (Fig. 1 B). Two constructs ending with residue 399 or 388 marked the transition from a δ-COP protein that was able to recognize R-based signals to one that was not (Fig. 1 B). We aligned the surrounding region of bovine, yeast, human, and fly δ-COP orthologues (Fig. 1 C) and found the region to be highly homologous, consistent with the evolutionarily conserved recognition of R-based sorting motifs (Michelsen et al., 2006). We deleted the conserved region (aa 388 to 413) from the full-length δ-COP Gal4-BD fusion construct and tested whether it had lost its ability to interact with the R-based sorting motif, which was indeed the case (Fig. 1 B). The internal deletion mutant of δ-COP had retained part of its function because δ-COP (Δ388-413) was still capable of binding β-COP in a two-hybrid assay (unpublished data).
β- and δ-COP are known to interact tightly in the trunk domain of the adaptor-like subcomplex (Faulstich et al., 1996; Eugster et al., 2000). To determine the region on the β-COP subunit required for the interaction with the R-based signal, we tested a series of constructs encoding N-terminally truncated forms of β-COP. Removing the N-terminal 210 aa had no effect on the specific interaction, whereas the signal was lost upon deletion of further 156 aa (Fig. 1 D). The responsible region was narrowed down to aa 318–338. A β-COP subunit lacking these residues was incapable of interacting with the strong R-based signal as assayed by the yeast two-hybrid system. As observed for δ-COP, alignment of the corresponding stretch with β-COP orthologues from yeast, human, mouse, and fly revealed high conservation of the putative binding site (Fig. 1 E). Our yeast two-hybrid analysis implicated the β- and δ-subunit of COPI as candidates harboring the binding site for R-based signals. This result could be explained by different hypotheses: (1) Both COPI subunits contain independent binding sites for the R-based signal; or (2) Both COPI subunits contribute to a common binding site and the affinity of the partial binding pocket is sufficient for a positive signal in the assay.
Two regions in β- and δ-COP are collectively required to recognize R-based signals in vivo
To distinguish between these possibilities, we created a strain that expresses a variant of the COPI coat specifically incapable of recognizing R-based sorting motifs. We exchanged the gene encoding β-COP against SEC26334DLD/NAN, (β-COP*) in the critical region and δ-COP against RET2Δ414-435 (δ-COP*). Strains expressing β-COP*, δ-COP*, or both were viable, although the double-mutant strain grew slowly and was temperature sensitive. These strains enabled us to test the localization of different reporters in the absence of the putative binding site(s).
We then assessed the functional consequence of removing the two putative binding sites for R-based signals individually and in combination by determining the subcellular localization of Pmp2YFP reporters presenting different variants of the R-based signal (Fig. 2). In mammalian cells, the two tails containing the sequences KLRRRRI or NVRNRRK were shown to function respectively as a very strong and a weak R-based ER localization signal by two independent quantitative assays (see Fig. 4 in Zerangue et al., 2001). Although these results show that the CD4-GFP reporter exposing the weak signal was able to leave the ER, most of the protein localized to the ER in the steady state as assessed by the GFP staining pattern in HeLa cells (Fig. S3). Consistent with the strict conservation of the relative efficacy of R-based signals in yeast (Michelsen et al., 2006), Pmp2YFP- NVRNRRK was also localized in the ER at steady state in wild-type yeast cells (Fig. 2 A). In the strains expressing either β-COP* or δ-COP*, Pmp2YFP-NVRNRRK (exposing a weak R-based signal) was no longer retained in the ER and efficiently reached the vacuole (Fig. 2 A, arrows), whereas Pmp2YFP-KLRRRRI (exposing a strong R-based signal) was still maintained in the ER (completely in the strain expressing δ-COP* and substantially in the strain expressing β-COP*; Fig. 2 A). Vacuolar localization is an indication of ER exit and loss of R-based signal recognition because Pmp2 reporters with inactive R-based signals efficiently reach the vacuole (Michelsen et al., 2006). In the strain coexpressing the two altered COPI-subunits, both R-based signal-exposing reporters strongly accumulated in the vacuole as demonstrated by differential interference contrast and colocalization with the fluorescent dye FM4-64 (Fig. 2, A and B). FM4-64 is taken up from the plasma membrane and reaches the vacuole via the endocytic route (Vida and Emr, 1995). In contrast, no colocalization with the vacuolar marker was observed for Pmp2GFP-KKTN. We tested whether the observed changes in steady-state localization could be explained by a reduced affinity of the double-mutant COPI coat for the respective signal-containing tails. Binding assays with cytosol obtained from wild-type or double-mutant yeast cells revealed that the mutant COPI complex containing both β-COP* and δ-COP* recognized a KKTE signal at the C terminus of Mst27 (Sandmann et al., 2003) almost as efficiently as the wild-type coat, whereas binding to the last 36 aa of Kir6.2 containing the LRKR signal was strongly reduced (Fig. 2 C).
To address the specificity of the sorting defect (Fig. 2 B) we investigated the steady-state localization of two additional membrane proteins known to be retrieved by COPI. Rer1 is a retrieval receptor for ER membrane proteins that is dynamically localized to the Golgi apparatus by COPI (Sato et al., 2001). Emp47 is a cargo receptor that shuttles between the ER and the Golgi (Schroder et al., 1995; Sato and Nakano, 2002). The C-terminal KTKLL signal of Emp47 is thought to be recognized by the WD40 domain of β′-COP (Eugster et al., 2004). Indirect immunofluorescence and subcellular fractionation on sucrose gradients revealed similar steady-state localization patterns for both Rer1 and Emp47 in wild-type and double-mutant cells (Fig. 3, A and B). The punctate staining pattern is consistent with their accumulation in Golgi sub-compartments. The vacuolar protein carboxypeptidase Y (CPY) was found to float to the light fractions (labeled 1, 2 in Fig. 3 B) of the gradient for both strains, whereas the heavy fractions (labeled 7, 8, 9) of the gradient were enriched in the ER marker Kar2. Detection of Pmp2YFP-NVRNRRK (exposing a weak R-based signal) revealed that the reporter cofractionated with the ER marker Kar2 in the wild-type strain and partially redistributed to the lighter fractions in the strain coexpressing both, β-COP* and δ-COP*. This confirms the mislocalization of this reporter in the double-mutant strain. The most apparent defect in the double-mutant strain consisted of multiple vacuoles as observed by FM4-64 staining (Fig. 2 B) or indirect immunofluorescence of a vacuolar membrane protein (Fig. 3 A).
We conclude that the COPI coat containing both β-COP* and δ-COP* is specifically incapable of recognizing R-based signals. At the same time this mutant coat is still capable of sorting three other cargo proteins with high fidelity. The two independent or one common binding site(s) formed by β- and δ-COP are necessary and sufficient to recognize R-based peptide sorting motifs. Importantly, the region contributing to the recognition site for R-based signals is highly conserved across eukaryotic species, consistent with the fact that the signals are effective and follow the same consensus in yeast and mammalian cells. The effects of mutating or deleting the critical stretch in either β- or δ-COP were additive, e.g., individual mutations in either β- or δ-COP caused considerable ER exit of the Pmp2YFP reporter exposing a weak R-based signal, whereas the reporter exposing the strong signal remained mostly ER-localized (Fig. 2 A). When both mutations were combined, the less efficient signal was completely inactive as an ER localization motif and reporters presenting the stronger LRKR or KLRRRRI signals accumulated outside the ER. This additive behavior supports the notion that the regions in the two subunits contribute to a common binding site that is completely distinct from the binding site recognizing KKXX signals.
The binding site for R-based signals in the trunk of COPI is in the same region that recognizes YXXΦ signals in adaptor complexes
No high-resolution structure of the COPI coat complex is available, but sequence homology and partial structures suggest that the β-, δ-, γ-, and ζ-COP subcomplex is structurally similar to the clathrin-adaptor core complex, whereas the α-, β'-, and ε-COP subcomplex is thought to be functionally equivalent to clathrin (Schledzewski et al., 1999; Owen et al., 2004; Langer et al., 2007). To better understand the structural relationship between the stretches identified in β- and δ-COP, we built a homology model of the adaptor-like COPI subcomplex containing β-, δ-, γ-, and ζ-COP based on the crystal structure of the clathrin adaptor 1 core (Heldwein et al., 2004) (Fig. 4 A; compare the original structure shown as Fig. 4 B for reference). Our homology model is consistent with the idea that the relevant regions of β- and δ-COP come into close proximity and with the concept of one binding site for R-based signals at the subunit interface. Strikingly, this binding site in the adaptor-like trunk is structurally comparable to the site where YXXΦ endocytic motifs are recognized by clathrin adaptors (Fig. 4 B, compare red and blue regions in Fig. 4 A to black ellipse). This suggests that the general architecture of the trunk structure has evolved to accommodate completely different cargo-sorting signals. It will be interesting to test how the recognition of R-based signals is coupled to the sorting of ion channel and receptor subunits.
Materials And Methods
Construction of plasmids
Molecular cloning followed standard procedures as described in Ausubel et al. (1997). All plasmids are listed in Table S1 . All constructs were verified by sequencing. Pmp2GFP and YFP fusions were as described in Michelsen et al. (2006). Two additional constructs were prepared where Pmp2 was fused to YFP followed by a unique NotI site. The NotI site encoded three alanines linking YFP and the tail. Oligonucleotides coding for the different tails were cloned into NotI and XhoI sites, resulting in constructs pR889 –AAATLASKLRRRRISLS and pR890 –AAATLASNVRNRRKSLS (compare Table S1).
The GST fusions are derivatives of the GST-MST27-KKXX construct (Sandmann et al., 2003). The construct is a fusion of the C-terminal tail of Mst27 (codons 159–234 including stop codon) to GST. The different R-based signals were introduced as oligonucleotides using a ClaI site (codon 226 of MST27) and an XhoI site present in the pGex6P vector. The sequences of the different tails are as follows in the indicated constructs (compare Table S1): pO732 -227DALLKKTE and pF291 -227DLLDALTLASSRGPLRKRSVAVAKAKPKFSISPDSLSGSRSHHHHHH. Bold-print letters indicate the peptide-sorting motif.
Construction of ret2 and sec26 mutant strains
The yeast strains used in this study are listed in Table S2 . A diploid heterozygous ret2Δ strain (Y25865) was transformed with an URA3-containing CEN plasmid (pQ849) carrying the temperature-sensitive ret2-1 allele. The strains were then sporulated and spores were selected for their capacity to grow on geneticin (the ret2Δ genotype was marked with a kanMX4 selection marker in strain Y25865) and for 5-fluororotic acid (5′FOA) sensitivity to ensure loss of the wild-type RET2 allele present in the diploid strain. Plasmids carrying either wild-type RET2 (Y25865-KM1) or RET2Δ414-435 (Y25865-KM2) under the control of the RET2 promoter were then integrated into the HIS3 locus of the resulting haploid strains. These strains were selected on 5-FOA for loss of the ret2-1-carrying URA3 plasmid. In Y25865-KM2, the presence of only the altered RET2 variant after gene swapping was confirmed by the polymerase chain reaction and by Western blot analysis of total cell lysates using an anti-COPI antiserum.
To test whether different mutant SEC26 alleles could substitute for the wild-type gene, we used a strain (RDY122) with an integrated copy of SEC26 under the control of the GAL1 promoter (Duden et al., 1994). The strain was transformed with a CEN plasmid carrying the respective variant of SEC26 (SEC26 334DLD/NAN or 325DILR/NAAA or Δ318-338) under the control of a MET25 promoter (p894-p897) and a LEU2 marker. Serial dilutions of individual transformants were spotted on selective medium lacking leucine and containing glucose to suppress transcription from the GAL1 promoter. All transformants that had received SEC26 wild- type or mutant alleles but not those that had received only the empty plasmid were able to grow in the presence of glucose indicating that all Sec26-derived proteins were functional components of coatomer.
Strain Y25865-KM2 was transformed with a URA3-marked 2μ plasmid carrying the wild-type SEC26 gene. Next, a natMX4 cassette disrupting the coding region of SEC26 was introduced by homologous recombination and the resulting transformants were selected for growth on nourseothricin-containing medium. This strain was then transformed with a LEU2-marked CEN plasmid carrying the different SEC26 alleles (SEC26 334DLD/NAN or 325DILR/NAAA or Δ318-338). Transformants were selected on 5′FOA- containing medium to select against the plasmid carrying the wild-type SEC26 gene yielding strains Y25865-KM3 and Y25865-KM4.
Large-scale transformation and microscopy
Large-scale transformations were performed in 96-well plates using the lithium acetate method (Ausubel et al., 1997) and transformants were plated on synthetic complete (SC) medium lacking uracil or leucine. Liquid cultures were inoculated in 96-well plates containing SC medium lacking uracil or leucine and viewed by epifluorescence after overnight growth at 30°C. For screening of the tetO7 promoter allele library 40 μg/ml doxycycline was added to the medium during the overnight growth period before microscopy. Automated microscopy of the two yeast libraries was performed on an Olympus ScanR screening system (40 × 0.95 objective) and 96-well glass-bottom plates (MMI).
Yeast two-hybrid analysis
Double transformants of strain AH109 containing Gal4-BD and -AD plasmid were selected on SC medium lacking tryptophane and leucine. In contrast to Zerangue et al. (2001), we replaced the auto-activating Gal4-BD fusions of bovine β- and ζ-COP with the respective fusions to the corresponding yeast COP subunits (encoded by SEC26 and RET3) and performed the two-hybrid assays at low stringency (e.g., dilution series of double transformants grown in SD lacking tryptophane and leucine were spotted on SD medium lacking tryptophane, leucine, and histidine to test for the activation of the HIS3 reporter gene). The plasmid encoding the Gal4-BD fusion to Ret3 proved toxic to the yeast and was thus omitted from further experiments. The deletion analysis was performed on bovine δ-COP and yeast β-COP.
Preparation of cell lysates
Crude cell extracts were prepared essentially as described by Schultz (1999). In brief, yeast cells were grown to logarithmic phase, harvested, and resuspended in cold wash buffer (20 mM Hepes, pH 7.4, 0.7 M sorbitol, and 1 mM PMSF). Cells in wash buffer were transferred to a syringe and pelleted for 5 min at 2,000 rpm. Supernatant was removed and the paste was extruded from the syringe into liquid nitrogen. Frozen cells were ruptured with a pestle and mortar filled with liquid nitrogen. The resulting white powder was transferred to a reaction tube and an equal volume of binding buffer (20 mM Hepes, pH 6.8, 2% glycerol, 150 mM KAc, 5 mM Mg(Ac)2, 1 mM EDTA, 1 mM DTT, and 0.1% Triton-X 100 containing protease inhibitor mix: 2.5 μg/ml leupeptin, 1.5 μg/ml antipain, 0.5 μg/ml chymostatin, and 1 μg/ml pepstatin A) was added. After thawing, the crude extract was cleared for 5 min at 2,000 rpm to remove unbroken cells.
Purification of GST fusion proteins
GST fusion proteins were induced in BL21 (DE3 star) carrying pRosetta by adding 0.2 mM IPTG for 2 h at 37°C. Cells were harvested and sonified in breaking buffer (20 mM Hepes, pH 6.8, 2% glycerol, 150 mM KAc, 5 mM Mg(Ac)2, 1 mM EDTA, 1 mM DTT, 1 mM PMSF, and 1× Complete protease inhibitor cocktail [Roche]). The lysate was cleared for 5 min at 2,000 rpm. The resulting supernatant was subjected to ultra-centrifugation for 30 min at 100,000 g. The GST fusion proteins were purified from the supernatant by incubating with GSH-agarose. The agarose column was washed with breaking buffer and proteins were eluted by adding elution buffer (20 mM Hepes, pH 9.5, 10 mM GSH, 5% glycerol, 150 mM KAc, 5 mM Mg(Ac)2, 1 mM EDTA, and 1 mM DTT). For purification of the GST fusion protein with −LRKR-HexaHis C-terminus a Protino-2000 Ni-column (Macherey) was used as per the manufacturer's instructions. Appropriate fractions were dialyzed overnight in binding buffer and 0.5 μg dialyzed protein bound to 1 μl GSH-agarose (Sigma-Aldrich).
For binding experiments, crude extract was cleared of all insoluble material for 30 min at 100,000 g and the supernatant was used for binding experiments. 100 μg total protein in 500 μl binding buffer (containing protease inhibitors, see above) was used for a binding assay with 2.5 μg bait on 5 μl GSH-agarose and incubated 2 h at 4°C. Beads were washed five times in binding buffer. Bound proteins were eluted with 1× SDS loading buffer containing 1 mM DTT, resolved by SDS PAGE, and detected by Western blotting using rabbit anti-COPI and rabbit anti-GST (Sigma- Aldrich) primary antibodies and an anti–rabbit AlexaFluor 680 (Invitrogen) secondary antibody. Blots were scanned (700 nm) and quantified with the Li-COR Odyssey system.
Subcellular fractionation on sucrose gradients
Separation of organelles was performed according to Nass and Rao (1998). In brief, cells were grown to mid-logarithmic phase, converted to spheroplasts, and lysed by douncing in a hypotonic buffer (0.3 M sorbitol and 50 mM triethanol amine, pH 8.9). The homogenate was layered on freshly prepared ten-step sucrose gradients and centrifuged for 12 h at 23,500 rpm in an SW28 rotor (Beckman Coulter). Membranes from the different fractions were pelleted by centrifugation (SW28, for 30 min at 23,500 rpm).
Immunostaining of yeast cells
Cells were grown to early logarithmic phase and fixed at room temperature in freshly prepared fixative for at least 8 h and processed as described by Roberts et al. (1991). In brief, spheroblasts were permeabilized by a short incubation in 1% SDS in 1.2 M sorbitol and then stained with the indicated primary and Alexa594-conjugated secondary antibodies (Invitrogen).
Microscopy was performed using a microscope (DM IRE2; Leica) controlled by OpenLab software (Improvision) with a 100×/1.4–0.7 HCX PL APO CS oil immersion objective. Images were captured by an Orca-ER CCD camera (Hamamatsu) with excitation at 470/40 nm and emission at 525/50 nm (GFP, YFP), or excitation at 510/40 nm and emission at 610 nm/longpass (FM4-64, Alexa594). All images shown are representative of several independent experiments. Steady-state staining with FM4-64 was performed in SC medium for 1 h at room temperature. For live-cell imaging, yeast were incubated in SC medium at room temperature. Fixed yeast cells were mounted in ProLong Gold AntiFade reagent with DAPI (Invitrogen). Live cell images of transiently transfected HeLa cells were acquired using a 63×/1.4–1.6 HCX PL APO oil immersion objective in L-15 medium (Leibovitz; Sigma-Aldrich) at 37°C. Images were transferred to Adobe Photoshop CS2 for slight gamma adjustments.
Comparative modeling of yeast COPI
A comparative three-dimensional model for the tetrameric subcomplex of yeast COPI was built based on the Crystal Structure of the clathrin adaptor protein core AP-1 (PDB: 1w63; Heldwein et al., 2004). Template identification for all four chains of the yeast COPI trunk lacking the appendage domains was performed by searching the PDB database (Berman et al., 2000) for suitable template structures in SwissModel Workspace (Arnold et al., 2006). Target-template alignments were generated by structure-guided multiple sequence alignment using 3DCoffee (O'Sullivan et al., 2004) for aligning γ-COP with template chain 1w63:A, β-COP with 1w63:B, and ζ-COP with 1w63:S. The alignment between δ-COP with 1w63:M was generated by HMM-HMM alignment (Soding, 2005). Model coordinates for the tetrameric complex were generated using satisfaction of spatial restraints (Sali and Blundell, 1993) after visual assessment of placements of insertions and deletions in the alignments.
Online supplemental material
Figure S1 shows microscopic images of selected hits in the reverse genetic screen expressing Pmp2YFP-LRKR and Pmp2YFP-KKTN. Figure S2 shows the localization of different Pmp2-GFP reporters in strains lacking ε-COP or expressing mutant α-COP. Figure S3 shows the steady-state localization of CD4GFP reporters exposing a very strong and a weak R-based signal in HeLa cells. Table S1 provides information about the plasmids and Table S2 about the yeast strains used in this study.
We thank Rainer Pepperkok for use of the EMBL Advanced Light Microscopy Facility and helpful advice and members of the Schwappach lab for discussions and careful reading of the manuscript. We are grateful to Rainer Duden for providing a yeast strain and a plasmid and to Hans Dieter Schmitt for the gift of Emp47-, Rer1-, and Ret2-antiserum. We are indebted to Julien Béthune, Britta Brügger, and Felix Wieland for continuous discussion and support throughout the project. The technical help of Martina Götzmann is gratefully acknowledged.
B. Schwappach was supported by the SFB638 of the Deutsche Forschungsgemeinschaft, the Landesstiftung Baden-Württemberg, the EMBO Young Investigator Programme, and the ZMBH. A. Spang was supported by the University of Basel.
U. Liebel's present address is Institute of Toxicology and Genetics, Forschungszentrum Karlsruhe, D-76021 Karlsruhe, Germany.
Abbreviations used in this paper: AD, activation domain; BD, binding domain.