Fluorescent Proteins as Proteomic Probes*S

Protein complexes mediate the majority of cellular processes. Knowledge of the localization and composition of such complexes provides key insights into their functions. Although green fluorescent protein (GFP) has been widely applied for in vivo visualization of proteins, it has been relatively little used as a tool for the isolation of protein complexes. Here we describe the use of the standard GFP tag to both visualize proteins in living cells and capture their interactions via a simple immunoaffinity purification procedure. We applied this method to the analysis of a variety of endogenous protein complexes from different eukaryotic cells. We show that efficient isolations can be achieved in 5–60 min. This rapid purification helps preserve protein complexes close to their original state in the cell and minimizes nonspecific interactions. Given the wide use and availability of GFP-tagged protein reagents, the present method should greatly facilitate the elucidation of many cellular processes.

Protein complexes mediate the majority of cellular processes. Knowledge of the localization and composition of such complexes provides key insights into their functions. Although green fluorescent protein (GFP) has been widely applied for in vivo visualization of proteins, it has been relatively little used as a tool for the isolation of protein complexes. Here we describe the use of the standard GFP tag to both visualize proteins in living cells and capture their interactions via a simple immunoaffinity purification procedure. We applied this method to the analysis of a variety of endogenous protein complexes from different eukaryotic cells. We show that efficient isolations can be achieved in 5-60 min. This rapid purification helps preserve protein complexes close to their original state in the cell and minimizes nonspecific interactions. Given the wide use and availability of GFP-tagged protein reagents, the present method should greatly facilitate the elucidation of many cellular processes. Molecular & Cellular Proteomics 4:1933Proteomics 4: -1941Proteomics 4: , 2005.
Protein complexes mediate the majority of cellular processes. Localization of such complexes provides key insights into their functions. There are a variety of tools for localizing individual proteins in space and time in the cellular milieu (1). Among the most widely used is the green fluorescent protein (GFP) 1 (1,2), which provides a convenient in vivo tag for following the dynamics of gene expression and protein localization (3). For example, GFP was recently used to genomically tag more than 4000 proteins for a large scale analysis of protein localization in Saccharomyces cerevisiae (4).
As well as localization, information on the composition of protein complexes provides key functional insights. Thus, a variety of methods have been devised for the study of protein interactions (5). Of these, immunoaffinity purification provides an effective means for isolating protein complexes and elucidating their composition (6). Immunoisolation is achieved with antibodies directed either 1) specifically against the proteins of interest or 2) against tags that are coupled to the proteins of interest. In the first of these strategies, production of high quality antibodies for each protein of interest can be timeconsuming and expensive. Moreover the efficacy of this approach depends strongly on the specificity and affinity of the antibody, which in turn may depend on the antigenicity of the protein in question. The second strategy, utilizing tagged proteins, is attractive because a single highly optimized antibody can serve as the immunoaffinity reagent. Commonly used tags include the FLAG and MYC peptides and Protein A (7). Recently the use of multiple tags, such as the two-step tandem affinity purification tag strategy, has become popular (8). Although all these tags have proved highly effective for the isolation of protein complexes, they cannot be readily visualized in living cells. Although GFP has the potential to serve as both a visualization and isolation tag, it has only occasionally been used for immunoaffinity purifications (9 -15, 49). One highly attractive option would be a simple, rapid procedure that can be applied to any GFP-tagged construct. Here we describe how the use of GFP to both visualize proteins in live cells and capture their interactions via immunoaffinity purification on magnetic beads promises to facilitate our understanding of the temporal and spatial dynamics of protein interactions.

MATERIALS AND METHODS
GFP Reagents-GFP was in-house cloned into the pGex 4T-3 vector, which contains the GST open reading frame, expressed from an inducible tac promoter, and immediately upstream of a thrombin cleavage site (Invitrogen). The GFP open reading frame was inserted in-frame between the SalI and NotI restriction enzyme sites in the multiple cloning sequence of this vector to create a single GSTthrombin cleavage site-GFP open reading frame. The resulting expression vector was freshly transformed into BL21 cells before each preparation. Two-liter cultures were grown at 37°C in LB medium (Fisher) with ampicillin until A 600 reached ϳ0.5. Cultures were then transferred to 30°C and induced with 0.5 mM isopropyl ␤-D-thiogalactoside (Stratagene) for 3.5 h. Cells were harvested, and pellets were resuspended in lysis buffer (PBS, 2 mM EDTA, 1:200 protease inhibitors (16) to make up to a total of 200 ml). Cells were lysed using an M110S microfluidizer (Microfluidics International Corp., Newton, MA). Alternatively lysis could be performed by French press or other methods. Cell debris were removed by centrifugation in an SS-34 swinging bucket rotor (Sorvall centrifuge RC-5B) at 10,000 rpm for 30 min at 4°C. A glutathione-Sepharose column was prepared for the purification of the GST-GFP fusion protein. Glutathione-Sepharose (2.5-ml bed volume) was added to a small column using a peristaltic pump and washed with 200 ml of lysis buffer. Sample was loaded onto the resin with two discrete passes through the column at 5 ml/min flow rate. The resin was washed with 250 ml of lysis buffer with protease inhibitors and 250 ml without protease inhibitors and allowed to drip to almost dry. The column was sealed, and 10 ml of lysis buffer, 1 mM DTT, and 100 units of thrombin (Amersham Biosciences) were added. The sample was cleaved from the resin by an overnight incubation at 4°C with gentle rotation after which the eluted protein was drained from the resin. The resin was then washed with a further 10 ml of lysis solution, and the wash was combined with the first elution. The sample was dialyzed against PBS, 5% glycerol and concentrated to 1 mg/ml (Centricon). This sample was submitted to Covance Immunology Services (Denver, PA) for the preparation of anti-GFP antibodies.
Anti-GFP Polyclonal Antibodies and IgG-A custom high titer anti-GFP polyclonal antibody was prepared at Covance using the in-house prepared GFP (825 g/animal) to inject female Elite rabbits. Rabbits, selected to have no preimmune activity against the cells of interest, were injected with 250 g of purified GFP with Freund's Complete Adjuvant (Covance). The production of antibodies was boosted and tested in the following steps: day 21, boost with 125 g of GFP with Freund's Incomplete Adjuvant (FIA, Covance); day 31, test bleed (ϳ5 ml of serum); day 42, boost with 125 g of GFP with FIA; day 52, test Bleed (ϳ5 ml of serum); day 63, boost with 125 g of GFP with FIA; day 73, production bleed (ϳ20 ml of serum); day 84, boost with 100 g of GFP with FIA; day 94, production bleed (ϳ20 ml of serum); day 105, boost with 100 g of GFP with FIA; day 115, production bleed (ϳ20 ml of serum); and day 118, terminal bleed (ϳ50 ml of serum). The anti-GFP polyclonal antibodies were then affinity-purified. A CNBr-activated Sepharose 4B resin (Amersham Biosciences) was conjugated to purified GFP using the manufacturer's instructions. Purified antibody was prepared using this resin following standard antibody affinity purification procedures (17). For comparison, rabbit polyclonal antibody to GFP was also purchased from Novus Biologicals. Purified rabbit IgG was purchased from ICN Pharmaceutical.
Conjugation of Magnetic Beads-Anti-GFP polyclonal antibodies or IgG were coupled to M-270 epoxy Dynabeads (Dynal) using an optimized version of the protocol suggested by the manufacturer. Briefly, magnetic beads were washed twice with 1 ml of 0.1 M sodium phosphate buffer, pH 7.4, with a 10-min period of mixing between the washes. The beads were resuspended with anti-GFP antibodies or IgG (2.5-10 g of Ab/1 mg of beads) followed by 0.1 M sodium phosphate buffer to bring the total volume to 13.3 l/1 mg of beads. After gentle mixing, 6.7 l of 3 M ammonium sulfate/1 mg of beads was added to give a total volume of ϳ20 l/1 mg of beads. Conjugation was carried out overnight on a rotor at 30°C. Saturation of 1 mg of beads was achieved with ϳ8 g of antibody or IgG. Following conjugation, coupled magnetic beads were washed sequentially with 0.1 M sodium phosphate buffer, pH 7.4; 100 mM glycine-HCl, pH 2.5; 10 mM Tris, pH 8.8; freshly prepared 100 mM triethylamine solution; four washes with PBS; one 15-min wash with PBS, 0.5% Triton X-100; and one wash with PBS, and then stored at 4°C in PBS, 0.02% NaN 3 . Beads could be stored for up to 1 month after which time their efficiency for isolation decreased by ϳ40%.
Yeast Strains-Yeast strains expressing chromosomal fusion of the DNA fragment encoding the IgG binding domain of protein A to Kap123 and of the GFP to Kap123 and Mcm2 were a kind gift from Benjamin Timney and Vincent Archambault from the Laboratory of Cellular and Structural Biology at The Rockefeller University. Yeast strains expressing chromosomal fusion of GFP to Apl1, Nup84, Cdc14, and Arp2 were purchased from Invitrogen. Cell culture chemicals were purchased from BD Biosciences.
Cell Imaging-Light and fluorescence microscopy images were acquired using the DeltaVision Image Restoration microscope (Applied Precision/Olympus).
Cell Disruption and Extraction-Yeast strains were grown to ϳ2.0 ϫ 10 7 cells/ml, harvested, and frozen as described previously (16). HeLa cells were grown to ϳ90% confluence, harvested in a minimal volume of PBS by gentle scraping with a rubber policeman, and frozen in liquid nitrogen. Cell lysis (both yeast and HeLa) was performed by cryogenic grinding using the Retsch MM 301 Mixer Mill (Retsch, Newtown, PA). Grinding was performed in five steps of 3min at 30 Hz inside 25-ml jars using 20-mm tungsten carbide grinding balls. The jars were cooled in liquid nitrogen between each step. An ϳ90% efficiency of cell disruption was confirmed by light microscopy. Lysis buffers were optimized individually for each studied protein complex (optimal conditions are given in Supplemental Table I).
Lysis buffer was added to the frozen cell powder (5 ml of buffer/g of cells), and the resulting cell lysate was homogenized for 10 s with a PT 10-35 Polytron (Kinematica), then slowly rotated at 4°C for 10 -15 min, and centrifuged for 10 min at 3000 rpm at 4°C. The resulting soluble fraction was used for the affinity purification experiments. Extraction efficiency was tested by Western blot analysis comparing the pellet and soluble fractions. Immunoaffinity Purification-Magnetic beads, coated with anti-GFP Ab and stored at 4°C in PBS, 0.02% NaN 3 , were washed three times with lysis buffer and added to the soluble fraction of cell lysate. Immunoaffinity purifications were achieved by slow mixing at 4°C. The magnetic beads were then collected using a Dynal magnet and washed rapidly six times with lysis buffer. The isolated protein complex was eluted from the beads for 20 min at room temperature in a fresh aqueous 0.5 N NH 4 OH, 0.5 mM EDTA solution. The resulting supernatant was frozen on liquid nitrogen and left to dry overnight by vacuum centrifugation. The pellet was suspended in SDS-PAGE sample buffer, separated on a 4 -12% NuPAGE Novex bis-Tris precast one-dimensional electrophoresis gel (Invitrogen) according to the manufacturer's specifications and stained with Coomassie Blue stain (Pierce) compatible with mass spectrometry.

Strategy for Rapid Isolation of Protein Complexes Containing a GFP-tagged Protein-
We used a single fluorescent tag to both visualize a given protein in live cells and to efficiently isolate it together with its stable interacting macromolecular partners. The method (Fig. 1) has the following sequential steps: 1) The protein of interest is tagged with GFP and visualized by fluorescence microscopy. 2) Parallel with visualization, the cells are frozen in liquid nitrogen. 3) The frozen cells are cryogenically lysed to help maintain protein complexes in their original state. 4) The GFP-tagged protein together with associated macromolecules is isolated on magnetic beads coated with polyclonal anti-GFP antibody. Isolation conditions are optimized to give efficient recovery of the tagged protein while maintaining its native interactions. 5) The isolated proteins are eluted from the magnetic beads and resolved by one-dimensional gel electrophoresis (one-dimensional SDS-PAGE). 6) The purified proteins are identified by MALDI mass spectrometry. If desired, the identified proteins can in turn be tagged with GFP, and the cycle can be repeated. We have found this procedure to be equally efficient with cyan fluorescent protein, yellow fluorescent protein (YFP), and other variants of GFP.
To render this procedure rapid, efficient, and robust, we carefully optimized each of the steps listed above. The success of the procedure is dependent on availability of a high affinity anti-GFP antibody. Hence we raised polyclonal antisera by immunizing rabbits with a high purity full-length GFP preparation (see "Materials and Methods"); prebleeds from these rabbits had no significant nonspecific activity, and the final production bleed sera titered on Western blots to better than 1 in 10 6 . Affinity-purified antibodies were prepared from these antisera using the same full-length GFP. Conjugation of the anti-GFP antibodies to magnetic beads was optimized to achieve saturation. The optimal amounts and concentrations of antibody and beads as well as the temperature and duration of the conjugation reaction are described under "Materials and Methods." Briefly saturation of 1 mg of magnetic beads was achieved with ϳ8 g of affinity-purified anti-GFP antibody in a total volume of 20 l (Supplemental Fig. S1).
During the immunoisolation step, the yield of the isolated tagged protein (and any associated macromolecules) is highly dependent on the conditions used for extraction and purification. Thus, we recommend optimization of extraction and purification conditions individually for each different protein complex of interest. The efficiency of the immunoisolation on the beads is routinely Ͼ90%, and the recovery of this material after the high pH elution step is Ͼ95% (e.g. Fig. 2B). Controls with anti-GFP beads incubated in lysate from wild type yeast and yeast expressing GFP gave no significant protein signal or only GFP, respectively; similar results were obtained with wild type HeLa cells and HeLa cells expressing GFP (plus a small vimentin contaminant in the latter).
Because of the amounts needed for immunoisolations, it is preferable to generate anti-GFP antibodies in-house as described under "Materials and Methods." However, in addition to our in-house generated polyclonal anti-GFP antibody, we tested several commercially available monoclonal and polyclonal anti-GFP antibodies for their utility in protein isolation; one of these proved suitable for isolating a tagged protein but with an increased level of nonspecific binding (see "Materials and Methods" and Supplemental Fig. S2).
Visualization and Isolation of GFP-tagged Protein Complexes-We tested the applicability of this method for the analysis of numerous protein complexes located in different subcellular compartments (e.g. cytoplasm, nucleus, nucleolus, mitochondria, and bud neck) (Fig. 2). Hence in parallel with their in vivo visualization, GFP-tagged Apl1, Arp2, and Kap123 (expressed at endogenous levels) were isolated with their interacting partners from yeast whole cell extracts. Similarly GFP-tagged Nup37 and Nup43 were purified from human cell extracts. A full list of the identified interacting proteins and the conditions found optimal for these immunoaffinity purifications is given in Supplemental Table I. AP-2R, the clathrin-associated protein complex (23), is a well characterized complex and was therefore selected as an appropriate case study for our methods ( Fig. 2A). Apl1, the ␤-adaptin of the large subunit of the AP-2R complex (23), is estimated to be present at ϳ1600 molecules per haploid cell (24). Yeast cells with GFP-tagged Apl1 expressed at endogenous levels were cultured; a small portion was examined by fluorescence microscopy, confirming the previously reported localization of Apl1 in vivo to the bud neck, while the rest was processed for immunopurification of Apl1-GFP with its associated macromolecular complex ( Fig. 2A). Our method recovered Apl1-GFP in high yield (Fig. 2B). Protein identification by mass spectrometry confirmed the co-purification of Apl1-GFP with all the known components of the AP-2R complex: Apl3, Apm4, and Aps2 (25,26). A schematic illustration of this complex, modified from (26), is shown in Fig. 2A. Additionally co-purified were Ede1 and Pmd1. Ede1 is the yeast homologue to human Eps15, which is involved in clathrin-mediated endocytosis, and was suggested to be involved in bud site selection (27,28). Pmd1 is a low abundance protein (24) reported to be a negative regulator of sporulation (29). The MS analysis-derived expectation value and protein sequence coverage for each of the purified proteins are presented in the Supplemental Table II. Another well characterized assembly is the Arp2/3 complex, involved in actin-based mitochondrial motility and maintenance of mitochondrial morphology, of which the actinrelated protein Arp2 is a member. Our method co-purified the known members of the Arp2/3 complex (30): Arp2, Arp3, Arc35, Arc40, Arc19, Arc18, Arc15, and other associated proteins, including Ede1, Bni1, Crn1, and Actin, only some of

Fluorescent Proteins as Proteomic Probes
which were previously reported to interact with Arp2 (31).
The nucleocytoplasmic transport factor Kap123 is an abundant protein involved in the import of ribosomal protein cargoes into the nucleus via the nuclear pore complex (NPC) (32), and so provides an example of a protein that interacts both with cytoplasmic proteins (its ribosomal import cargoes) and nuclear proteins (the FG repeat nucleoporins in the nuclear pore complex). As expected, the proteins that co-purified with Kap123 were indeed ribosomal protein cargo (Rps1b (32)), FG repeat nucleoporins from the nuclear pore complex (Nup60, Nsp1, Nup82, Nup116, and Nup2) (32), and Kap60, which is known to associate with Nup2 (33).
We also tested the efficacy of our anti-GFP immunopurifications in vertebrate cells. For this study, we chose two components (Nup37 and Nup43) of the human Nup107-Nup160 subcomplex, a major structural building block of the nuclear pore complex (9,18,34,35). Moreover if our methods are robust and reciprocal, these two proteins should co-purify with the same set of proteins including each other; indeed this proved to be the case (Fig. 2C).
Effects of Incubation Time on Protein Recovery and Nonspecific Binding-During optimization of our immunoaffinity purification protocol, we observed that the length of time used for incubation of the lysate with the GFP-Ab beads had profound effects both on the efficiency and specificity of isolation of protein complexes. These effects are illustrated for the isolation of a previously well studied subcomplex from the yeast NPC, i.e. the Nup84 subcomplex that is known to consist of the proteins Nup133, Nup120, Nup145C, Nup85, Nup84, Seh1, and Sec13 (36 -39). GFP-tagged Nup84 was isolated together with its associated proteins at a series of different incubation times (Fig. 3). At each time point, we purified all seven expected components of the Nup84 subcomplex with the amount of this subcomplex reaching a maximum after ϳ1 h; the major co-purifying proteins are identical to those originally reported to be present in the complex when isolated using the Protein A tag (36,37). In addition, we observed interactions of the Nup84 complex with several other members of the NPC including the nucleoporins Nup188, Nup192, Nup157, Nup170, Pom152, Nic96, Nup145N, Nup59, Nup53, Nup57, and Nsp1 (39). Certain of these interactions (i.e. those between the Nup84 subcomplex and Nup157, Nup170, Nic96, and Nup145N) have been re-ported recently (40). Significantly the interactions that we observed between the Nup84 subcomplex and other members of the NPC were stable only at the shorter incubation times; the amount of these associated nucleoporins maximizes at ϳ1 h and thereafter falls off steeply with time, indicating that their interactions with the Nup84 subcomplex are weaker than those holding the components of the Nup84 subcomplex together. Another important advantage of short incubation times is in minimizing nonspecific binding as illustrated for immunopurifications of Nup84-GFP (Fig. 3) and Mcm2-GFP (47,48) (Supplemental Fig. S3). Such nonspecific binding, typically of highly abundant proteins, is a common phenomenon in immunoisolation studies (30,31,41). Tests indicate that the bulk of the nonspecific binding occurs as a result of binding to the components of the affinity-purified protein complex (data not shown). In summary, the very high affinity of the anti-GFP reagent allows the use of short incubation times (Յ1 h), which are advantageous both for preserving the integrity of endogenous complexes and for maximizing the amounts of the isolated complexes while minimizing the amounts of contaminants.
One concern with using short purification steps (5-30 min) is the resulting lowered yield of the tagged protein complex (Fig. 3). Therefore, we investigated the consequences of increasing amounts of anti-GFP beads during the purification, assaying both immunoisolation efficiency and nonspecific binding (Supplemental Fig. S4). The efficiency of isolation improved with the increase of resin amount up to a saturation point. The amount of resin necessary to reach saturation varies according to the amount of tagged protein in the cell lysate (i.e. tagged protein abundance). Nonspecific binding increased only slightly with resin amount when short incubation times were used for purification but increased significantly with longer incubation times. We conclude that short immunoisolation times and increased amounts of anti-GFP beads are optimal.
Profiling Cell Cycle-dependent Protein Complexes-Two significant challenges for any proteomic method are the study of cell state-dependent interactions and the study of interactions involving low abundance proteins. The S. cerevisiae phosphatase Cdc14, a key regulator of mitotic exit (42), exemplifies both challenges as it forms cell cycle-dependent interactions with proteins of low abundance. Here we en- riched for interactions specific to particular cell cycle states by comparing isolations of GFP-tagged Cdc14 (expressed at endogenous levels) from 1) asynchronous cells and 2) cells arrested in the early stages of S phase using hydroxyurea (43).
Immediately prior to immunopurification, representative cells from the two populations were examined by fluorescence microscopy (Fig. 4). Upon immunopurification from asynchronous cells, GFP-tagged Cdc14 co-isolated with several proteins with which it is known to interact (including several components of the spindle pole body and known substrates of Cdc14 during mitosis (44)) as well as other proteins associated with the cell cycle that were not observed previously in immunoisolations of Cdc14 (e.g. Cnm67 and Cdc31) (Fig. 4A). Some of these proteins are present in low copy numbers in the cell, including Sli15, Nud1, Spc110, and Rad9 (24).
Following the addition of hydroxyurea, Cdc14 co-purified with most of the proteins observed in the isolates from asynchronous yeast cells (Fig. 4B). Notably more Net1 co-purified with the arrested cells, consistent with its role for the Net1-Cdc14 interaction in retaining Cdc14 in the nucleolus prior to mitosis (45). Additionally we observed a reduction in amount of the co-isolating spindle pole body proteins such as Nud1, Cnm67, and Cdc31.
Other proteins were found to be significantly enriched in immunopurifications from hydroxyurea-arrested cells compared with asynchronous cells; these include Sir2, Bnr1, Est1, Rad24, Ctf18, and Ctr86. Sir2 (a component of the chromatinsilencing complex) is known to interact with Cdc14; Bnr1 and Est1 are cell cycle-regulated proteins; and both Rad24 and Ctf18 are required in the cell cycle, the former as a checkpoint protein and the latter for sister chromatid cohesion. These proteins are known or suspected key players in cell cycle regulation and so are potential substrates for the Cdc14 phosphatase. Although elucidation of the functional significance of these interactions clearly requires additional experiments, our results illustrate how dynamic and rare interactions may be visualized and identified by our anti-GFP immunopurification methods.

DISCUSSION
This study reports a robust strategy that uses a single, fluorescent tag to both visualize proteins in live cells and efficiently isolate their interacting macromolecular partners. We have demonstrated the applicability of this method to the analysis of protein complexes located in various subcellular compartments and have isolated even low abundance proteins in both yeast and mammalian systems. We show that high affinity anti-GFP antibodies permit rapid, single step immunoisolations, which in turn minimize nonspecific binding and maximize recovery of transiently interacting partners.
Clearly these GFP-based immunopurifications are at least as efficient as those using Protein A. However, the use of GFP as an affinity tag has significant advantages. First, the GFP tag can be both visualized and purified, allowing us to simultaneously combine information on the cellular localization of a protein with information on its interaction partners in the same cell population. Second, the expression and visualization of GFP fusion proteins does not appear to be species-dependent. Similarly our immunopurification protocol is not speciesdependent as we have already immunopurified GFP-tagged proteins from viruses, bacteria, yeast, mammalian tissue culture cells, and mice. Third, there already exist a vast number of strains in which GFP has already been incorporated into a particular protein almost exclusively for the purpose of visualizing the protein within cells, tissues, or organisms. Moreover the antibody is equally efficient with cyan fluorescent protein, YFP, and other variants of GFP; by immunoblot analysis the antibody appears to have equally high titers on all these variants and has been used for immunoprecipitation of protein complexes on several variants of YFP (data not shown). Our protocol now allows the researcher to investigate the interactions made by each of the tagged proteins in all these strains. Given this wide use and availability of GFPtagged protein reagents, the present method should greatly facilitate the elucidation of many cellular processes.