A Universal Strategy for Proteomic Studies of SUMO and Other Ubiquitin-like Modifiers*S

Post-translational modification by the conjugation of small ubiquitin-like modifiers is an essential mechanism to affect protein function. Currently, only a limited number of substrates are known for most of these modifiers, thus limiting our knowledge of their role and relevance for cellular physiology. Here, we report the development of a universal strategy for proteomic studies of ubiquitin-like modifiers. This strategy involves the development of stable transfected cell lines expressing a double-tagged modifier under the control of a tightly negatively regulated promoter, the induction of the expression and conjugation of the tagged modifier to cellular proteins, the tandem affinity purification of the pool of proteins covalently modified by the tagged modifier, and the identification of the modified proteins by LC and MS. By applying this methodology to the proteomic analysis of SUMO-1 and SUMO-3, we determined that SUMO-1 and SUMO-3 are stable proteins exhibiting half-lives of over 20 h, demonstrated that sumoylation with both SUMO-1 and SUMO-3 is greatly stimulated by MG-132 and heat shock treatment, demonstrated the preferential usage of either SUMO-1 or SUMO-3 for some known SUMO substrates, and identified 122 putative SUMO substrates of which only 27 appeared to be modified by both SUMO-1 and SUMO-3. This limited overlapping in the subset of proteins modified by SUMO-1 and SUMO-3 supports that the SUMO paralogues are likely to be functionally distinct. Three of the novel putative SUMO substrates identified, namely the polypyrimidine tract-binding protein-associated splicing factor PSF, the structural microtubular component α-tubulin, and the GTP-binding nuclear protein Ran, were confirmed as authentic SUMO substrates. The application of this universal strategy to the identification of the pool of cellular substrates modified by other ubiquitin-like modifiers will dramatically increase our knowledge of the biological role of the different ubiquitin-like conjugations systems in the cell.

The post-translational modification of proteins provides the cell with the ability to mount a rapid response to external changes and stimuli. The best-characterized types of posttranslational modifications have been those involving the conjugation of small chemical groups to the target protein, such as phosphorylation and acetylation. However, during the last few years the post-translational modification of proteins by the covalent conjugation of small proteins has gained relevance as a very important mechanism to affect protein function. This is best exemplified by the conjugation of polyubiquitin chains to a target protein, leading to the proteasomal degradation of the modified protein. Currently there are 11 known small protein modifiers, namely ubiquitin, ISG15, AUT7, APG12, NEDD8, the SUMO proteins (SUMO-1, -2, & -3), HUB1, FAT10, URM1, MNSF, and Ufm1, all of which are related to the prototypical member (ubiquitin) and are therefore considered to be ubiquitin-like proteins (1,2). Conjugation with these modifiers exerts a wide variety of effects on the target protein, including changes in protein conformation, activity, protein-protein interactions, and cellular localization. This diversity of effects is associated with the large and chemically varied surface provided by these modifiers.
The best-characterized ubiquitin-like modifiers are ubiquitin itself and the SUMO proteins. SUMO was independently discovered by three groups during yeast two-hybrid screens for partners to the promyelocytic leukemia (PML) 1 protein (3), Rad51/Rad52 (4), and the Fas/APO-1 death domain (5). Because of its multiple discovery, the modifier initially had several early designations including Ubl1, PIC1, and sentrin. Sequence comparisons suggested that Ulb1/PIC1/sentrin was the mammalian homolog of the Saccharomyces cerevisiae SMT3 gene, an essential gene in S. cerevisiae previously identified in a screen for suppressors of a yeast temperaturesensitive MIF2 gene (6,7). While the biological functions of this newly identified mammalian protein were unknown, it appeared to be a member of the ubiquitin family. These initial reports were rapidly followed by the discovery that the Ran GTPase-activating protein, RanGAP1, was covalently modified by conjugation of this same protein, now designated as SUMO (8,9). A subsequent study determined that SUMO was conjugated to RanGAP1 via an isopeptide bond between the carboxyl group of SUMO glycine 97 and the ⑀-amino group of RanGAP1 lysine 526 (10), confirming that SUMO not only shared sequence relatedness to ubiquitin, but also was conjugated to substrates in a chemically analogous fashion. However, the SUMO-conjugating enzyme, Ubc9, was shown to function only with SUMO and not with ubiquitin, demonstrating that these modification pathways are biochemically parallel yet distinct (11).
The pathway of SUMO conjugation exemplifies the conjugation pathway used for all the known ubiquitin-like protein modifiers. Briefly, SUMO is synthesized as an inactive molecule that must be cleaved in order to expose the di-glycine motif used for conjugation. This is accomplished by the action of a class of cysteine proteases, termed SUMO proteases. Upon cleavage, SUMO is activated in an ATP-dependent process by the dimeric structure formed by the SUMO-activating enzymes SAE1 and SAE2. SUMO is then transferred from SAE1/SAE2 to an internal cysteine residue in the SUMOconjugating enzyme Ubc9. Finally, Ubc9 conjugates SUMO to the ⑀-amino group in a lysine residue located in the target protein, forming an isopeptide linkage. This final step is enhanced by proteins known as SUMO-ligases, which accelerate the transfer of SUMO to the target and are thought to provide specificity to the conjugation system by regulating the interaction between the target and the conjugating enzyme (12)(13)(14). Once a protein has been sumoylated, it can be desumoylated again by the action of the SUMO proteases. To date, three different types of SUMO ligases and six different SUMO proteases have been identified in mammals (14). Interestingly, even though the biochemical pathway of SUMO conjugation and deconjugation is well defined, the regulatory mechanisms that determine the specificity and extent of SUMO conjugation in the cell remain mostly unknown.
SUMO modification exerts a large variety of effects on its targets, altering their cellular localization, stability, ability to interact with other proteins, and activity, which can be either stimulated or repressed (14). For instance, many of the known SUMO substrates are transcription factors, and while for most of them SUMO modification decreases their transcriptional activation function (15)(16)(17)(18)(19)(20), for others sumoylation augments their activity (21)(22)(23). A wide range of cellular processes are currently known to be affected or regulated by sumoylation, including chromosomal organization and function, DNA repair, nuclear transport, and signal transduction pathways. Obviously, the types of cellular processes regulated by sumoylation are determined by the identity of the proteins targeted by SUMO conjugation. A broad identification of the spectrum of proteins modified by sumoylation is required to better define the range of cellular events regulated by sumoy-lation and is likely to provide significant clues about the mechanisms that provide specificity to the system. Similarly, defining the spectrum of proteins modified by any given ubiquitinlike modifier is essential to our understanding of the range of cellular processes affected by each ubiquitin-like modifier and the mechanisms that dictate their specificity.
Clearly, proteomics studies defining the range of proteins targeted by every ubiquitin-like modifier could provide great insights into the cellular role and regulation of each conjugation system. Discovery of entire proteomes is a very challenging task, but the identification and characterization of posttranslational modifications on a proteomic scale is an even more difficult one, as for any given protein the amount of modified protein is only a small fraction of the total cellular pool, and a single protein may be modified at multiple sites. The compartmentalization or subfractionation of proteomes makes the analysis of the sample and the interpretation of the data more practical (24,25). Recently, several groups have performed proteomics studies aimed at defining the range of cellular proteins targeted by sumoylation (26 -30), following the lead established by an earlier proteomics study on ubiquitin conjugation (24). The most successful studies providing the most extensive lists of novel substrates for SUMO (27,29) were performed with the yeast S. cerevisiae, as this system can be easily scaled-up, providing virtually unlimited amounts of starting material, and can be easily manipulated to replace the endogenous SUMO gene with one coding for tagged versions of the modifier. The studies performed with mammalian cell lines have yielded a much more limited spectrum of novel potential SUMO targets (26,28,30), due in part to the difficulties inherent to the production of sufficient quantities of starting material. However, both the yeast and mammalian studies were limited by the apparent low specificity provided by the use of single-stage affinity purification methods for the enrichment of the sumoylated proteins. Single-stage affinity purifications, primarily those based on interactions between charged chemical groups and specific amino acids in the target proteins, but also including those based on proteinprotein interactions such as antigen-antibody-based affinity purifications, are known to produce relatively high backgrounds of spurious interacting proteins (31,32).
In this article, we present a strategy for enriching and identifying SUMO-modified proteins in mammalian cell lines that is applicable to the identification of the pool of proteins modified by any other ubiquitin-like modifier and therefore represents a universal strategy for proteomics studies of ubiquitin-like modifiers. The overall strategy involves the development of a stable transfected cell line expressing a double-tagged SUMO under a tightly negatively regulated promoter, followed by the induction of the expression and conjugation of the tagged modifier to cellular proteins, the use of a tandem affinity purification (TAP) method for the specific enrichment of the modified proteins, and the identification of the enriched proteins by LC-MALDI-MS/MS. The application of this strat-egy allowed us to evaluate several basic aspects of SUMO biology (such as its half-life, the effects of SUMO overexpression on the cell cycle, and the changes in overall sumoylation induced by different environmental stresses), allowed us to compare the array of substrates modified by SUMO-1 and SUMO-3, and led to the identification of 122 putative SUMO substrates, some of which had been previously defined as genuine SUMO targets. Three of the novel potential SUMO substrates identified, namely the polypyrimidine tract-binding protein-associated splicing factor PSF, the GTP-binding nuclear protein Ran, and the structural microtubular component ␣-tubulin, were confirmed as bona fide SUMO substrates by immunoblotting or in vitro sumoylation reactions, further supporting a role for SUMO in transcriptional regulation, RNA processing, nuclear transport, and maintenance of chromosomal stability, and suggesting a novel role for SUMO in the regulation of cellular microtubular structures. The application of this proteomics approach to the identification of the pool of cellular substrates modified by other ubiquitin-like proteins could dramatically increase our knowledge of the physiology and regulation of the ubiquitin-like conjugation systems in the cell.

Development of Stable Transfected Cell Lines and Flow
Cytometry-All cells lines used in these studies were grown in complete medium containing 1ϫ Dulbecco's modified Eagle medium (DMEM) supplemented with 10% fetal bovine serum (Gemini Bio-Products, Woodland, CA), in a humidified incubator at 37°C, 5% CO 2 , unless otherwise indicated. To develop stable transfected cell lines expressing tagged SUMO proteins, the genes encoding the human SUMO-1 and SUMO-3 genes (accession numbers NP003343 and AAH08420, respectively) were inserted into the pBAC-2cp vector (Novagen, Inc., Madison, WI), thereby adding a sequence coding for a hexa-histidine tag, a thrombin recognition sequence, the 15-aa residue S-tag, and an enterokinase recognition site to the 5Ј end of the genes. The tagged genes were PCR-amplified and cloned into the pcDNA5/FRT/TO vector (Invitrogen Corp., Carlsbad, CA), which contains a Flp recombination target (FRT) sequence. The derivative plasmids obtained were co-transfected with the Flp recombinase expression plasmid pOG44, using LipofectAMINE 2000, into the FlpIn T-REx HEK293 (F293) cell line (all from Invitrogen Corp.), a HEK293-derivative containing a single integrated FRT site. Cells maintaining an integrated copy of the transfected plasmid were selected in medium containing 10 g/ml of Blasticidin and 100 g/ml of Hygromycin. Upon selection, the isogenic cell populations were amplified and maintained in antibioticcontaining medium, tested for ␤-galactosidase activity (lost upon gene insertion at the FRT site), and several aliquots were frozen in liquid nitrogen for long-term storage. Expression of the His-S-SUMO proteins was induced by adding tetracycline (Tet) to the culture medium at a final concentration of 1 g/ml. For cell cycle distribution analyses, the cell were cultured with or without Tet, changing the medium every 24 h, for a total of 72 h. The cells were trypsinized, washed with 1ϫ PBS, fixed in 70% ethanol for 30 min, and stained for 10 min at 37°C with a solution containing 50 g/ml propidium iodide, 4 mM sodium citrate, 150 mM NaCl, 0.1% Triton X-100, and 30 U/ml RNase I, pH 7.8. Upon staining, the cells were maintained in the dark on ice and analyzed in a FACSCalibur flow cytometer using Cell Quest software (Becton Dickinson, San Jose, CA).
Protein and Transcript Stability Analyses-For pulse-chase experiments, the F293-SUMO cell lines were plated at 3 ϫ 10 6 cells per flask in 25-cm 2 flasks, Tet induced for 24 h, starved for 1 h, pulselabeled with 200 Ci of trans-35 S label (MP Biomedicals, Irvine, CA) for 1 h, washed and chased in unlabeled complete medium, and collected at different times post-chase. Both the starvation and pulse labeling were performed in Met (-) , Cys (-) 1ϫ DMEM supplemented with Tet. The cells were collected and processed for protein purification as described in "Tandem Affinity Purification" below. The purified samples were resolved by SDS-PAGE, and the band corresponding to free-SUMO was quantified by phosphordensitometry. The half-life of each protein was defined as the time at which half the initial counts were present in the purified free SUMO, as calculated from the values obtained above. For studies aimed at measuring transcript stability, the F293-SUMO cell lines were plated at 4 ϫ 10 6 cells per dish in 10-cm Petri dishes and induced with Tet. Twenty four hours postinduction, Actinomycin D (ActD) was added to the medium at 5 g/ml, and the cells were collected at different times post-ActD addition. RNAs were purified using the RNAqueous®-Midi kit (Ambion, Inc., Austin, TX) as described by the manufacturer, and the His-S-SUMO transcripts were detected by RT-PCR using primers targeting the sequence coding for the His-S-tag, thereby avoiding cross-detection of the endogenous SUMO transcripts and allowing the direct comparison of the His-S-SUMO-1 and His-S-SUMO-3 transcripts. For Northern blot analyses, the RNAs purified from the samples described above were run on a formaldehyde-agarose gel, transferred by capillary action using a TurboBlotter™ device (Schleicher & Schuell BioScience, Inc., Keene, NH) to a GeneScreen Plus® membrane (PerkinElmer Life Sciences, Inc., Boston, MA) and hybridized to a probe complementary to the His-S-tag, thus allowing the direct comparison of the tagged SUMO transcripts as indicated above for the RT-PCR analyses.
Tandem Affinity Purification (TAP)-For TAPs, the F293-SUMO cell lines or the parental F293 cell line were plated at 1 ϫ 10 7 cells in 175-cm 2 flasks and Tet induced 24 h after plating. To allow optimal His-S-SUMO expression and conjugation, the cells were maintained in the presence of Tet for 72 h, replacing the medium every 24 h. Eight hours before collection, the proteasomal inhibitor MG132 was added to the medium at a final concentration of 5 M, and 1 h before collection the cells were incubated at 41°C, 5% CO 2 . At the time of collection, the cells were washed in 1ϫ PBS (140 mM NaCl, 2.7 mM KCl, 10 mM Na 2 HPO 4 , 1.8 mM KH 2 PO 4 , pH 7.4), lysed in 1ϫ denaturing buffer A (8 M urea, 100 mM NaH 2 PO 4 , 10 mM Tris, 10 mM ␤-mercaptoethanol, pH 8.0) supplemented with 0.2% Triton X-100 and 20 mM N-ethylmaleimide (NEM), and the resulting extracts were either stored at Ϫ70°C or processed immediately for TAP. The extracts were sequentially passed through 21, 23, and 27.5 gauge needles, sonicated, and cleared by centrifugation at 15,000 ϫ g for 10 min at 4°C. The clarified extracts were incubated with His-Select™ Nickel Affinity Gel (Sigma-Aldrich Co., St. Louis, MO) in a circular rocker for 16 h at 4°C. After incubation, the resulting suspension was poured through an empty column and the beads were washed with 50 bead volumes of 1ϫ denaturing buffer A, and 20 bead volumes of 1ϫ denaturing buffer B (1 M urea, 100 mM NaH 2 PO 4 , 10 mM Tris, pH 6.3). The bound proteins were eluted with 5 bead volumes of 1ϫ denaturing buffer C (1 M urea, 100 mM NaH 2 PO 4 , 10 mM Tris, pH 3.9), and the eluate was neutralized with an equal volume of 1ϫ neutralizing buffer (100 mM NaH 2 PO 4 , 190 mM Tris, pH 8.8). The neutralized eluate was incubated with S-protein agarose (Novagen, Inc.) in a circular rocker for 16 h at 4°C, washed with 90 bead volumes of 1ϫ PBS, and the bound TAP-purified proteins were either eluted with 2 volumes of 4ϫ SDS-PAGE sample buffer (100 mM Tris, 20% glycerol, 8% SDS, 0.02% bromphenol blue, 4% ␤-mercaptoethanol) or by digestion with EKMax™ Enterokinase (Invitrogen Corp.) in 1ϫ enterokinase reaction buffer (50 mM Tris pH 8.0, 1 mM CaCl 2 ) for 16 h at 37°C.
In Vitro Sumoylation Assays-In vitro sumoylation assays were performed as previously reported (33). Briefly, 1 g of purified target protein was incubated with or without 1 g of SAE1/SAE2, 200 ng of Ubc9, and the indicated amounts of SUMO1, in a buffer containing 50 mM Tris pH 8.0, 5 mM MgCl 2 , 5 mM ATP, and 0.5 mM DTT, in a final volume of 25 l. The reactions were carried at 30°C for 90 min, stopped by the addition of 4ϫ SDS-PAGE sample buffer, boiled for 3 min, and processed for immunoblotting as described above.
LC-MALDI-MS/MS-Enterokinase-eluted TAP-purified proteins were digested with sequencing-grade tryspin (Promega Corp., Madison, WI) at a 100:1 protein:trypsin ratio and 37°C for 4 h. To maximize cleavage efficiency, the proteins were denatured for 20 min at 85°C, cooled down to 37°C, and incubated with another aliquot of trypsin overnight, as previously reported (34). The resulting solution (ϳ0.5 ml) was concentrated in a speed-vac, and the pellet was resuspended in 20 l of 2% ACN, 0.1% TFA. Next, 10 l of sample were injected onto a 150-m ϫ 10-cm column (Vydac) using an LC-Packings autosampler and pumps (LCPackings, Sunnyvale, CA). A gradient of 90 min from 2 to 60% ACN was used to elute the peptides from the column at 1 l/min. Then 5 mg/ml ␣-cyano-4hydroxycinnamic acid (Fluka, Buchs, Switzerland) were mixed with the eluant through a "T" junction (Upchurch Scientific, Oak Harbor, WA) at 1.8 l/min. The resulting mixture was spotted directly onto a MALDI plate using the LC-Packing Probot. Spots were obtained every 6 s. A total of 624 spots were obtained per plate, and typically two plates were obtained per injection. The spots were analyzed using an Applied Biosystems 4700 Proteomics Analyzer (Applied Biosystems, Framingham, MA). Initially each spot was analyzed in reflectron mode. The resulting spectra were analyzed, and the spots with the highest intensity level for each mass obtained were used to acquire MS/MS data. The MS/MS data were analyzed using GPS explorer (Applied Biosystems) and an in-house version of the MASCOT (www.matrixscience.com) search engine. Identified proteins from either the F293-SUMO-1 or the F293-SUMO-3 cell lines were only considered if they had a minimum of one peptide with an individual score greater than 44. Proteins with only one or two peptides, with at least one having a score greater than 44, were confirmed by de novo sequencing. For the control sample, the selection criterion was reduced to any protein identified, regardless of score. Any protein found in the control sample and the F293-SUMO cell lines was removed from the list of identified proteins.

RESULTS
Previous studies by ours (35,36) and several other groups have revealed a relatively long list of viral proteins modified by SUMO (reviewed in Ref. 37). While such studies have suggested different roles for the sumoylation of the viral proteins, a more thorough understanding of the interactions between viruses and the host sumoylation system requires monitoring the overall changes in sumoylation occurring during infection. To this end, we decided to develop a strategy for the broad identification of the spectrum of proteins modified by sumoylation. Furthermore, as at least one of the other ubiquitin-like modifiers [the interferon-stimulated gene ISG-15 modification system (38)] is known to be up-regulated by viral infection, we sought to create a general strategy applicable to proteomic studies of other ubiquitin-like modifiers as well.
For most proteins targeted by the sumoylation system, the apparent amount of sumoylated protein at any given time appears to represent only a small fraction of the total cellular pool (14). Therefore, the first requirement for the proteomic evaluation of the pool of cellular proteins modified by SUMO is to enrich the sumoylated proteins while excluding and minimizing the amount of unmodified proteins. This is best achieved by the use of TAP approaches. To make the strategy applicable to the identification of the pool of cellular proteins targeted by any ubiquitin-like modifier, a commercially available TAP tag was added to the N-terminal region of SUMO by placing the genes for SUMO-1 and SUMO-3 in the pBAC-2cp vector (Novagen, Inc). This procedure introduced a sequence coding for a (His) 6 ϳS-peptide TAP tag and an enterokinase recognition site upstream from the SUMO genes ( Fig. 1a). To prevent any undesirable effect due to the overexpression of SUMO, and to avoid the limitations associated to transient transfection approaches, stable transfected cell lines were developed using an inducible expression system for the controlled overexpression of the tagged SUMO. To this end, the tagged genes were cloned into a mammalian expression plasmid containing a Tet-regulated operator and an FRT sequence, and the resulting plasmids were transfected into a 293 human embryonic kidney cell line derivative containing a single integrated FRT site. The polyclonal populations of cells that maintained an integrated copy of the plasmids were antibiotic-selected to produce two isogenic cell lines dubbed F293-SUMO-1 and F293-SUMO-3. The expression of the (His) 6 ϳS-peptideϳSUMO protein (hereafter designated His-S-SUMO) in these cell lines was negatively regulated by the constitutively expressed Tet repressor gene (TetR), and turned on by the addition of Tet to the culture medium (  (Fig. 1c). Although intriguing, these slight differences do not seem directly associated with SUMO overexpression as Tet induction did not trigger further changes in cell cycle distribution, and His-S-SUMO expression in the absence of Tet was minimal. Therefore, SUMO overexpression did not seem to induce an overall increase in total sumoylation or to have gross deleterious or advantageous effects on cellular growth.
Interestingly, even though the tagged modifiers were expressed from the same promoter and in the same locus in the two cell lines developed, His-S-SUMO-3 consistently accumulated to higher levels than His-S-SUMO-1 (Fig. 1b), paral-leling differences previously reported between SUMO-1 and SUMO-3 in untransfected cells (39). Therefore, the cause of this difference was sought experimentally. The half-life of each His-S-SUMO was measured in pulse-chase experiments performed by purifying SUMO from metabolically labeled cells collected at different times post-chase, and was determined to be over 20 h, indicating that SUMO-1 and SUMO-3 are very stable proteins (Fig. 2a). Then, we measured the stability of the transcripts by RT-PCR using primers complementary to sequences located in the His-S tag, therefore enabling us to use the same set of primers for the detection of the His-S-SUMO-1 and the His-S-SUMO-3 transcripts and avoiding interferences due to the endogenous SUMO transcripts. The cells were Tet-induced for 20 h, treated with ActD, and collected at different times post-ActD treatment. The RT-PCR analysis indicated that the transcripts for His-S-SUMO-1 and His-S-SUMO-3 were both fairly stable as the intensity of the products remained constant up to 8 h post-ActD treatment. However, the His-S-SUMO-3 transcripts seemed more abundant than the His-S-SUMO-1 transcripts (Fig. 2b). This difference in abundance was confirmed by Northern blot analysis of the samples collected at different times post-ActD treat- ment, using a probe complementary to the tag (Fig. 2b). Therefore, the difference in the levels of His-S-SUMO protein expressed by the two F293-SUMO cell lines reflects differences in the accumulation of their respective transcripts, although the molecular basis for this remains unknown.
The next step was to standardize a procedure to consistently stimulate the conjugation of the overexpressed His-S-SUMOs. A previous report indicated that protein-damaging stimuli induce SUMO-2/3 conjugation (39). Therefore, we tested several different stress-inducers for their ability to increase the incorporation of the His-S-SUMO proteins into high-molecular-mass forms indicative of SUMO conjugation. Among several different conditions tested, an 8-h treatment with the proteasomal inhibitor MG-132 combined with a 1-h exposure at 41°C before harvesting led to the most substantial increase in the levels of expression and conjugation of His-S-SUMO-1 and His-S-SUMO-3 (Fig. 3a). This indicated that, unlike previously reported in Cos 7 cells (39), in the F293-SUMO cell lines stress-induced SUMO conjugation is not a property exclusive of SUMO-3 but instead is shared by SUMO-1. For all subsequent experiments, the F293-SUMO cell lines and the parental F293 cell line used as a control were induced using these conditions to ensure maximal conjugation of the His-S-SUMO.
Next, a TAP protocol was developed to purify the pool of sumoylated cellular proteins. TAPs minimize the background of co-purifying contaminant proteins (31) and therefore can potentially enhance the identification of low-abundance proteins by MS approaches. The covalent nature of the linkage between ubiquitin-like proteins and their targets, and the nature of the affinity tags selected, allowed the use of strong denaturing conditions during the initial stages of purification. Such conditions are expected to inactivate the de-conjugating enzymes, ensure proper solubilization of the modified targets [many of which are known to be trapped in insoluble nuclear domains (40 -42)], and disassemble protein complexes thereby preventing the co-purification of interacting proteins. Therefore, the induced cells were collected directly in a buffer containing 8 M urea and 0.2% Triton X-100. Interestingly, preliminary trials indicated that some desumoylation occurred during the affinity purification of the sumoylated proteins, even after the use of the strongly denaturing conditions indicated above. Therefore, NEM was incorporated as an essential component of the buffer used during cell lysis and urea (43)(44)(45)(46). The use of a low urea, low salt, and high pH buffer proved to yield the best recovery of tagged proteins, and was therefore incorporated into the TAP protocol. Application of the resulting TAP protocol (described in detail in "Experimental Procedures") led to a substantial enrichment of sumoylated proteins from the cell lines expressing the His-S-SUMOs, and a low background of untagged proteins from the parental cell line, as verified by SDS-PAGE and immunoblot analyses of aliquots taken at different stages of the purification process using anti-SUMO antibodies (Fig. 3, a and b) and Coomassie blue staining (Fig. 3c).
To further validate the TAP protocol, we tested by immunoblotting for the presence of several known SUMO targets in the affinity-purified samples. The SUMO targets RanGAP-1 (8 -10, 47), PML (42,48,49), p53 (50,51), and HSF-1 (21) were all successfully detected in TAP-purified samples, and their altered migration, indicative of sumoylation, further validated the TAP developed (Fig. 4, a-d). For RanGAP-1, a single band suggestive of a single SUMO-conjugation event per molecule was obtained, and the apparent amount of sumoylated protein recovered from the F293-SUMO-1 cells was significantly higher than the one recovered from the F293-SUMO-3 cells (Fig. 4a). As RanGAP-1 is preferentially modified at a single site by SUMO-1 (39), these observations strongly indicated that although the cells had been stimulated for over-sumoylation, the overall specificity of the sumoylation reaction had not been compromised in the F293-SUMO cell lines. Interestingly, PML and HSF-1 appeared preferentially modified by His-S-SUMO-3 (Fig. 4, b and c, respectively), whereas p53 appeared preferentially modified by His-S-SUMO-1 (Fig. 4d).
Next, a protocol for the proteomic analysis of the TAPpurified proteins was developed. Direct protein identification from SDS-PAGE gel bands was attempted with very limited success, probably due to the presence of large numbers of proteins in very small quantities in each band. Therefore, an alternative method was employed to allow the identification of the purified proteins. First, the proteins that remained bound to the S-protein beads after the TAP procedure were eluted off the beads by digestion with enterokinase. This digestion step added an additional degree of specificity to the purification, as only those proteins containing the enterokinase recognition sequence (which is contained within the His-S-tag) should be susceptible to enterokinase cleavage and release from the beads. Next, the enterokinase-released proteins   were digested with trypsin and the resulting peptides were resolved by HPLC and eluted directly onto MALDI plates. Finally, the samples spotted on the MALDI plates were analyzed by MALDI MS, first in reflectron mode and then the spots with the highest intensity level for each mass obtained were used to acquire MS/MS data. This approach allowed the execution of repeated analyses on any given spot whenever it was considered necessary. In total, we performed two independent TAP experiments for every cell line used. To maximize the total number of proteins identified and increase the confidence of such identifications, the raw data captured in each experiment were combined and analyzed together. Altogether, the use of the TAP method herein developed in conjunction with LC-MALDI-MS/MS analyses of the proteins released by enterokinase digestion (exemplified in Fig. 5, a  and b) resulted in the identification of 122 putative sumoylated proteins, including four proteins previously suggested to be sumoylated but for which no further validation as authentic SUMO substrates exists [namely actin (28,29), ataxin (28,52), the transcription intermediary factor 1-␤ (TIF1-␤) (26), and tubulin (29)], and three well-characterized SUMO targets (namely DNA topoisomerase I, histone H2A, and TFIIA) ( Table  I). None of the proteins identified by LC-MALDI-MS/MS analysis of the TAP-purified proteins from the F293 parental cell line (negative control) showed a best ion score above the threshold used for positive identification with the F293-SUMO TAP-purified samples. However, three proteins identified in the F293-SUMO TAP-purified samples were also identified in the negative control (although with a low ion score) and were therefore excluded from the list of putative sumoylated proteins (Table II). Among those was enterokinase, the enzyme used to release the TAP-purified proteins from the S-protein agarose beads, which was identified in the F293 parental, F293-SUMO-1, and F293-SUMO-3 TAP samples, as expected. Further details on the proteomic data obtained for every TAP-purified sample are presented in Supplemental  Tables I, II, and III. The proteomic data obtained is summarized in Fig. 5c. Most of the proteins identified were either transcription factors, nucleic acid binding proteins, or cellular structural components. Out of the total number of proteins identified, 62 were found exclusively in the F293-SUMO-1 cell line, 34 were found exclusively in the F293-SUMO-3 cell line, and the remaining 27 were found in both cell lines, therefore suggesting limited overlapping in the array of substrates modified by SUMO-1 and SUMO-3. Interestingly, none of the SUMO targets used to validate the TAP approach by immunoblotting were identified by LC-MALDI-MS/MS. This suggests that even though immunoblotting of TAP-purified sam- ples is impractical for large-scale identification of sumoylated proteins, it is perhaps the most-sensitive way to determine if a given protein is sumoylated in vivo.

TABLE I Putative SUMO-1 and SUMO-3 substrates identified by LC-MALDI/MS/MS analysis of TAP-purified proteins
Lastly, immunoblot analyses of TAP-purified samples were performed to confirm the sumoylation of three selected novel putative SUMO targets identified by LC-MALDI-MS/MS: the polypyrimidine tract-binding protein-associated splicing factor or PSF, the structural microtubular component ␣-tubulin, and the GTP-binding nuclear protein Ran. PSF was excluded from the final list of putative SUMO targets due to its presence in the TAP sample purified from the parental cell line (Table II). However, because the highest peptide scores and the number of peptides identified in the parental and the F293-SUMO-3 cell lines were substantially different, we considered it important to verify if PSF was an authentic SUMO substrate. The other two targets, Ran and ␣-tubulin, were selected because of their biological significance and the availability of specific antibodies. To verify the specificity of the purification used for testing the novel putative sumoylated targets, an immunoblot analysis was performed using anti-RanGAP-1 FIG. 6. PSF, ␣-tubulin, and Ran are authentic SUMO substrates. a-d, to confirm the SUMO modification of PSF, ␣-tubulin, and Ran, TAP-purified proteins were resolved in a 5-15% gradient SDS-PAGE gel, transferred to Immobilon™ membranes, and immunoblotted with antibodies directed against the specific proteins tested. a, RanGAP-1 (positive control used to verify the specificity achieved in the TAP used to validate the novel SUMO substrates). b, PSF. c, tubulin. d, Ran. e, in vitro sumoylation of GST-Ran. Bacterially expressed affinity-purified GST-Ran was sumoylated in vitro using affinity-purified SUMO-activating enzymes (Sae2/1), Ubc9, and increasing amounts of SUMO-1. The in vitro sumoylation reactions were denatured with sample buffer, boiled, resolved in a 10% SDS-PAGE gel, transferred to an Immobilon membrane, and immunoblotted with anti-GST antibodies. No significant sumoylation was observed in a GST-only control (data not shown), as previously reported by others (62). TCE, unfractionated cell extracts. TAP, TAP-purified proteins bound to the S-protein agarose at the end of the TAP procedure, eluted by treatment with SDS-PAGE sample buffer. P, parental F293 cell line. S1 and S3, Tet-induced heat-shocked and MG-132-stimulated F293-SUMO-1 and F293-SUMO-3 cell lines, respectively. Arrowheads and arrows indicate the position of the unmodified and sumoylated proteins, respectively. antibodies, as this protein had been previously established as an optimal positive control for the selective purification of sumoylated targets. Similar to our previous analysis, only the SUMO-modified forms of RanGAP-1 were detected in the TAP-purified samples (Fig. 6a, lanes 5 and 6), and no Ran-GAP-1 was detected in the TAP purification performed with the parental F293 cell line (Fig. 6a, lane 4), therefore indicating that the TAP purification worked successfully. The immunoblot analysis performed with the anti-PSF polyclonal antibody detected the unmodified form of PSF in the TAP-purified samples obtained from the parental and the SUMO-1 and SUMO-3 derivative cell lines. However, in the TAP-purified sample obtained from the F293-SUMO-3 cell line, two additional distinct bands exhibiting apparent molecular masses consistent with the addition of one or two chains of SUMO to PSF were also detected (Fig. 6b, lane 6). Such bands were not detected in the TAP samples from the parental or the F293-SUMO-1 cell lines (Fig. 6b, lanes 4 and 5). This finding suggests that PSF is an authentic SUMO target and that it is preferentially modified with SUMO-3. A similar analysis was performed using antibodies directed against ␣-tubulin. The anti-␣-tubulin monoclonal antibody produced a complex profile in the total cell extracts, but reacted primarily with a 50-kDa protein, corresponding to the expected molecular mass for ␣-tubulin (Fig. 6c, lanes 1-3). Interestingly, the only band detected in the purified samples was a 70-kDa band detected in the F293-SUMO-3-purified sample (Fig. 6c, lane  6). The altered migration of this band is within the range expected for a SUMO-modification-associated shift, and therefore strongly supports the hypothesis that ␣-tubulin is also a bona fide SUMO target. Lastly, a similar immunoblot performed with an anti-Ran monoclonal antibody detected Ran in the TAP-purified sample from the F293-SUMO-3 cell line (Fig. 6d, lane 6), but did not detect any Ran on the TAP samples from the parental or the SUMO-1 derivative cell lines (Fig. 6d, lanes 4 and 5). However, high-molecular-mass forms of Ran were detected exclusively in the total cell extracts and not in the TAP-purified samples, and only upon prolonged exposure of the immunoblot to the film. As the immunoblot analysis failed to provide conclusive evidence of the sumoylation of Ran, we performed an in vitro sumoylation experiment using affinity-purified bacterially expressed Ran, and purified sumoylation components, also from bacterial expression systems. The use of recombinant proteins expressed in bacteria guaranteed that any potentially sumoylated Ran product had to be produced during the in vitro sumoylation reaction and could not be due to spurious cross-reactivities with other cellular proteins. In the presence of a full set of sumoylation components, Ran was readily sumoylated, and the apparent concentration of the sumoylated form of Ran increased proportionally to the amount of free SUMO-1 used in the sumoylation reaction, therefore demonstrating that Ran is also an actual SUMO target (Fig. 6e). Altogether, the above demonstration that the tested proteins are legitimate SUMO targets strongly supports the hypothesis that the majority of the new sumoylation substrates identified by the proteomic analysis presented are also authentic SUMO targets.

DISCUSSION
This report presents the development and testing of a strategy for the assessment and identification of the pool of sumoylated proteins in a mammalian cell line. The overall strategy (summarized in Fig. 7) is clearly applicable to the proteomic analysis of the pool of proteins modified by any other ubiquitin-like modifier, and therefore represents a universal strategy for proteomic studies of ubiquitin-like modifiers. This methodology constitutes a major improvement over the use of transient transfection as a way to deliver the tagged modifier into the cells (26), single-step affinity purifications as a way to purify the pool of modified proteins (26 -30), and SDS-PAGE followed by in-gel digestion and peptide mass fingerprinting as a way to identify novel SUMO targets (26). The application of this method allowed us to study some of the biological characteristics of SUMO (i.e. protein stability, effects of SUMO overexpression on cell cycle, and stimulatory treatments for conjugation), indicated the preferential conju- gation of SUMO-1 or SUMO-3 to some known SUMO targets, and provided an extensive list of potential SUMO-1 and SUMO-3 sumoylation substrates in mammalian cell lines, three of which, namely PSF, Ran, and ␣-tubulin, were confirmed as bona fide SUMO substrates. Furthermore, this method will allow the evaluation of changes in the pool of modified proteins throughout the cell cycle, during cellular differentiation, and among different cell lines for SUMO and all other ubiquitin-like modifiers.
The stable transfected cell lines developed for this study expressed, in addition to the endogenous SUMOs, a His-Stagged version of SUMO-1 or SUMO-3. While the size of the tag used is significant in comparison to the size of SUMO (58-aa residues for the His-S-tag versus 97-or 93-aa residues for SUMO-1 or SUMO-3, respectively), and large tags may decrease the efficiency of conjugation (27), our data indicated that under the stress conditions employed (heat shock and MG-132 treatment) the tagged SUMOs were efficiently conjugated to cellular targets. In fact, contrary to our expectations, heat-shocking the cells in coordination with exposure to MG-132 significantly increased the conjugation with both His-S-SUMO-1 and His-S-SUMO-3, and not exclusively with His-S-SUMO-3. This suggests that SUMO conjugation as a whole, and not only with SUMO-2/3 as previously indicated (39), may be included among the cellular responses against stress. The increased conjugation observed under such conditions does not seem to alter the specificity of the sumoylation system because, although the overall sumoylation is substantially increased, the sumoylation profile observed for the SUMO substrates tested appeared mostly unchanged as compared with studies by other groups and by direct comparison with the parental cell line. Therefore, the increased conjugation achieved by heat shock and MG-132 treatment likely resulted in an overall increase in the fraction of sumoylated versus unsumoylated forms for most SUMO targets, and not in an overall change in the subset of sumoylated cellular proteins. While the molecular basis for the stimulatory effect mediated by MG-132 is unknown, it suggests a connection between proteasomal degradation, ubiquitination, and sumoylation. This connection is unlikely to involve proteasomal degradation of SUMO itself, as SUMO was shown to have a long life in the cell. The relatively long life of SUMO suggests that SUMO conjugation is more likely to be regulated at the conjugation and de-conjugation stages (by controlling the abundance and/or the activity of the SUMO ligases and isopeptidases) rather than by regulating the overall abundance of SUMO in the cell, as in the absence of rapid SUMO turn-over it would be hard to quickly decrease the abundance of SUMO in the cell. In fact, in the absence of a stimulatory signal for conjugation most of the tagged SUMO remained unconjugated even when it was overexpressed, further stressing the relevance of factors affecting the conjugation and de-conjugation stages. Therefore, we postulate that MG-132 may act by stabilizing SUMO-E3 ligases required for the efficient conjugation of SUMO to a wide array of cellular substrates, while the heat shock may up-regulate those SUMO-E3 ligases and perhaps down-regulate some of the SUMO isopeptidases.
The observed differences in transcript abundance between His-S-SUMO-1 and His-S-SUMO-3 are particularly difficult to explain in view of the fact that both were under the control of the same promoter region, had the same upstream and downstream regulatory sequences, and had the same chromosomal location. However, such differences correlated well with the observed apparent abundance of each SUMO in the cell lines developed. It remains to be investigated if similar differences in transcript abundance exist between the endogenous SUMO-1 and SUMO-3 transcripts. Obviously, although little attention has been given to the mechanisms regulating SUMO expression, detailed knowledge of such mechanisms is essential to our overall understanding of sumoylation and its physiological roles in the cell. Interestingly, upon heat shock and MG-132 induction the differences observed in the expression of His-S-SUMO-1 and His-S-SUMO-3 disappeared. In fact, more potential SUMO substrates were identified in the F293-SUMO-1 samples than in the F293-SUMO-3 samples, and only ϳ23% of all the proteins identified were found in both groups of samples. This provides further support to the observation that a certain degree of specificity allowing the preferential modification of specific substrates with each SUMO modifier remains even when the SUMO modifiers are overexpressed and overconjugated. This specificity also supports the hypotheses that the SUMO paralogues are functionally different, therefore providing different properties to the target protein, and that the selection of a SUMO paralogue for the modification of a given target is likely to be specifically regulated.
Proteomic studies face two interrelated issues: 1) lack of specificity, leading to the identification of proteins fortuitously purified due to the nature of the contaminant protein and the purification strategy employed; and 2) lack of sensitivity, leading to a failure to identify some of the proteins present in the sample under analysis. The purification method presented here for the enrichment of proteins modified by ubiquitin-like modifiers substantially decreases the likelihood that any given protein identified by the proteomics analysis may be a spurious contaminant, as it involves two affinity purification steps, each based on a different type of intermolecular interaction, plus the final sequence-specific release of the purified proteins by enterokinase. In fact, in sharp contrast with other previous SUMO-proteomics studies (28,53), very few proteins were identified in our negative control (consisting of the TAP-purified sample obtained from the parental cell line, presented in Table II and Supplemental Table III), leading to the exclusion of a small subset of proteins from the final list of putative SUMO substrates. Even more, one of the excluded proteins, PSF, was shown to be an authentic SUMO substrate by immunoblotting analyses. However, albeit minimal, spuri-ous co-purifications are still probable, so the possibility remains that some of the novel putative SUMO targets identified may not be bona fide SUMO substrates. None of the proteins identified by immunoblotting during the validation of the TAP protocol were identified by LC-MALDI-MS/MS analysis of the TAP-purified proteins under the stringent identification criteria applied for protein identification. This indicates that the list of putative SUMO substrates presented in Table I still represents only a limited fraction of all authentic SUMO substrates. Furthermore, it also indicates that the approach used to identify the purified proteins could be modified to increase its sensitivity. Out of all the steps involved in the TAP described, the one that is likely to result in the largest losses is the enterokinase treatment, as it is hard to experimentally assess the recovery achieved and optimize the conditions employed in this step. An attractive experimental alternative to the use of enterokinase for the release of the modified proteins from the S-beads is the use of peptidases specific for the ubiquitin-like modifier under study. This would potentially lead to a further increase in the specificity of the method, decrease the signal contributed by the modifier (as the modifier would remain bound to the beads), and consequently increase the sensitivity of the method for the proteomic detection of the cellular proteins targeted by the ubiquitin-like modifier. This alternative is currently under evaluation in our laboratories by the use of the SUMO-specific Ulp1 protease.
Out of the putative SUMO substrates presented herein, three novel SUMO substrates were confirmed as bona fide SUMO targets: ␣-tubulin, PSF, and Ran. ␣-Tubulin was identified as a putative SUMO substrate in a previous proteomic study (29), but it was not formally validated as a genuine SUMO target. ␣-Tubulin, in conjunction with ␤-tubulin, forms the structural unit of cellular microtubules. Mammalian microtubules originate in the centrosome and in coordination with kinesin and dyneins form the molecular motors responsible for the intracellular transport of vesicles and organelles and the migration of chromosomes during mitosis and meiosis. Tubulins are subjected to several types of post-translational modifications, including detyrosination, acetylation, phosphorylation, palmitoylation, polyglutamylation, and polyglycylation. Some of these post-translational modifications affect the interaction between microtubules and motor proteins (54). While tubulin sumoylation may play a similar role, it is tempting to speculate that sumoylation may confer to this cytoskeletal component the ability to concentrate or sequester other sumoylated proteins at specific intracellular locations, acting as "nucleation" sites to increase the local concentration of specific proteins, therefore enhancing the formation of multimeric protein machines. Interestingly, although only the sumoylation of ␣-tubulin was confirmed, ␤-tubulin was also identified as a putative SUMO-1 and SUMO-3 target. Furthermore, actin, the other major cytoskeletal component and the structural unit of microfilaments, was identified as a putative SUMO substrate here as well as in a previous SUMO pro-teomic study (28). Should all of these cytoskeleton components be authentic SUMO substrates, sumoylation could be a major regulator of cellular architecture and transport. Alternatively, sumoylation of these cytoskeleton structural units could be restricted to specific stages of the cell cycle and therefore could be relevant only for a limited set of processes, such as chromosomal segregation. Clearly, further studies are required to explore these alternatives.
PSF and Ran, the other two novel SUMO substrates that were validated in this study, were not identified as potential SUMO targets in any of the previous SUMO proteomic studies. PSF was initially characterized as a factor that co-purified with the polypyrimidine tract-binding protein (PTB) and appeared essential for pre-mRNA splicing, hence it was named PTB-associated splicing factor (or PSF). PSF exhibits a varied cellular distribution that includes the nucleolus, the nuclear membrane, the nucleoplasm, and a novel nuclear domain termed paraspeckles. Functionally, PSF appears to be a multifunctional protein with DNA and RNA binding properties, and has been implicated in a variety of cellular processes including pre-mRNA splicing, intranuclear retention of promiscuously edited RNAs, transcriptional repression, and enhancement of the helicase activity of DNA topoisomerase I (55). Sumoylation could clearly regulate any of the activities suggested above for PSF. Ran is perhaps the most intriguing novel substrate presented. Unlike PSF and tubulin, Ran lacks a predicted consensus sumoylation site on its sequence. However, a recent report indicated that the frequency at which SUMO is added to Lys residues located in nonconsensus sequences is much higher than previously recognized (56). Therefore, Ran sumoylation is likely to involve a noncanonical target sequence that, if defined, could reveal structural features common to other SUMO targets. Ran plays three major roles in the cell, acting as a regulator of nuclear transport, spindle assembly, and post-mitotic nuclear envelope assembly. In all of these roles, a Ran-GTP gradient is used to direct spatially regulated processes in reference to chromosomal localization. The Ran-GTP gradient is established via the localization of the guanine nucleotide exchange factor RCC1, the GTPase-activating protein RanGAP-1, and the Ran-binding proteins RanBP1, 2, and 3 (57,58). Sumoylation is already known to play a role in these processes as only the sumoylated form of RanGAP-1 binds to RanBP2 (8,9,47), and RanGAP-1 remains bound to RanBP2 throughout mitosis (59). Furthermore, sumoylation appears to regulate the nuclear traffic of several of the known SUMO substrates, as recently reviewed (60,61). Ran sumoylation may provide yet another mechanism to control the cellular events regulated by Ran, therefore increasing the relevance of SUMO for those cellular events. It would be interesting to determine if the proportion of sumoylated Ran increases at specific stages during the cell cycle, particularly during mitosis, as this may indicate if Ran sumoylation is equally relevant to all of the specific roles attributed to Ran. Altogether, the three novel SUMO substrates herein presented open new and exciting areas of SUMO research that will require extensive exploration.
Sumoylation is likely to be a rather transient modification as the fraction of sumoylated versus unsumoylated forms for any given protein is very small. This makes the proteomics evaluation of the total pool of sumoylated proteins intrinsically difficult. The same is probably true for all other ubiquitin-like modifiers. The proteomics approach herein presented allows for the rapid and highly specific enrichment of the pool of proteins modified by a given ubiquitin-like modifier, and, combined with state-of-the-art MS techniques, it allows for the rapid identification of the cellular targets modified by the ubiquitin-like modifier under study, as supported by the proteomics data presented for SUMO-1 and SUMO-3. The application of this proteomics approach to the identification of the pool of cellular substrates modified by other ubiquitin-like proteins could dramatically increase our knowledge of the physiology and regulation of the ubiquitin-like conjugation systems in the cell.