Specification of SUMO1- and SUMO2-interacting Motifs*

SUMO proteins are ubiquitin-related modifiers implicated in the regulation of gene transcription, cell cycle, DNA repair, and protein localization. The molecular mechanisms by which the sumoylation of target proteins regulates diverse cellular functions remain poorly understood. Here we report isolation and characterization of SUMO1- and SUMO2-binding motifs. Using yeast two-hybrid system, bioinformatics, and NMR spectroscopy we define a common SUMO-interacting motif (SIM) and map its binding surfaces on SUMO1 and SUMO2. This motif forms a β-strand that could bind in parallel or antiparallel orientation to the β2-strand of SUMO due to the environment of the hydrophobic core. A negative charge imposed by a stretch of neighboring acidic amino acids and/or phosphorylated serine residues determines its specificity in binding to distinct SUMO paralogues and can modulate the spatial orientation of SUMO-SIM interactions.

SUMO proteins are small ubiquitin (Ub) 5 -related modifiers that become conjugated to cellular substrates and regulate diverse cellular processes including cell cycle progression, intracellular trafficking, transcription, and DNA repair (1)(2)(3). Like Ub, a SUMO protein is covalently attached to target proteins through an isopeptide bond by a mechanism similar to that of ubiquitination, which involves E1, E2, and E3 enzymes (4). In mammals, three SUMO paralogues are commonly expressed: SUMO1 shares about 45% identity to SUMO2 and SUMO3, while SUMO2 and SUMO3 are 96% identical to each other (2,5).
The structures of all three SUMO paralogues resemble the globular and compact Ub-like fold (6,7). The differences of SUMO1 and SUMO2 are mostly found in the second ␤-strand and the ␣-helix of both proteins (7). In cells, different SUMO paralogues appear to share common properties but also have some distinct functions. For example, the promyelocytic leukemia protein is conjugated to all three SUMO paralogs (8,9), whereas RanGAP1 is preferentially modified with SUMO1 (10) and topoisomerase II with SUMO2/3 during mitosis (11). Furthermore, the distribution of the SUMO paralogues within cells seems to be different. SUMO1 is uniquely found within the nucleoli, the nuclear envelope, and cytoplasmic foci, whereas SUMO2/3 are accrued on chromosomes at an earlier point in the nuclear reformation process (12). Interestingly, there is a larger pool of free, non-conjugated SUMO2/3 than of SUMO1 (10).
In addition to targeting different substrate proteins, the functional properties of SUMO isoforms in vivo might also reflect their ability to mediate distinct protein-protein interactions. Indeed, recent studies have shown that SUMO paralogues can promote non-covalent binding to other proteins containing specific motifs that recognize SUMO paralogues. Minty and coworkers defined a Ser-Xaa-Ser motif surrounded by hydrophobic and acidic amino acids as a SUMO-interacting motif (SIM) (13). Biophysical studies of the SIM in PIAS revealed that the small hydrophobic region is an essential determinant of SUMO recognition (14). Moreover, the SUMO-binding motif was proposed as Lys-Xaa [3][4][5] 2 in yeast proteins (15). Recent publications revealed that the hydrophobic core can bind both parallel and antiparallel to SUMO (16,17). It is thought that sumoylated targets may control cell functions depending on their ability to interact with effectors containing SUMO-binding motifs. However, most of the SUMO interacting studies were done with the SUMO1 paralogue.

-[Val/Ile]-[Ile/Leu] 2 -Xaa 3 -[Asp/Glu/Gln/Asn]-[Asp/Glu]
In this study we describe the identification and characterization of novel SUMO1-and SUMO2-binding partners containing a universal SIM. The molecular and structural details are presented explaining the basis for SIM binding to distinct SUMO paralogues. We show that the E3 ligase PIASx␣ is phosphorylated in vivo within the SIM and that phosphorylation influences its binding to SUMO1 but not to SUMO2. In the case of TTRAP, a protein that binds SUMO2 much stronger than SUMO1, and PIASx␣ we show that negative charged amino acids surrounding the hydrophobic core influence binding to SUMO1 but not to SUMO2.

EXPERIMENTAL PROCEDURES
Yeast Two-hybrid Screen-Sequences corresponding to SUMO1(⌬GG) and SUMO2(⌬GG) were subcloned in pYTH9 vector between SalI and BglII restriction site creating fusion proteins with Gal4-DNA-binding domain. Both vectors were introduced using lithium acetate/polyethylene glycol transfomation with herring testis carrier DNA into Y190 yeast strain cDNA libraries were then similarily introduced and transformed cells were grown on agar plates containing a synthetic dropout medium (BD Bioscience) without leucine, tryptophan, histidine, and with 25 mM 3-amino-1,2,4-triazole. Colonies that grew on the selection medium were transferred to a filter and assayed for ␤-galactosidase activity with substrate X-gal (5-bromo-4-chloro-3-indolyl-␤D-galactopyranoside).
Plasmid DNA was extracted using a glass bead disruption method and were amplified by transformation and lysis of DH5␣ bacteria. Plasmids were then retransformed into yeast containing the bait to confirm binding and grown on the same agar plates as described above. After X-gal test plasmids were sequenced.
To compare the strength of interaction between SUMO1 and SUMO2, 1.5 g of each plasmid were retransformed in yeast containing SUMO1 and SUMO2 in parallel and grown on the agar plates described above. After 6 days three different colonies (if colonies grew at all) were replicated to another agar plate. After 3 days colonies were transferred to a filter to compare the growth of yeast containing SUMO1 and SUMO2 and the interacting partners.
Acidic deletion mutants were constructed using site-directed mutagenesis. Sequences of the primers are available upon request. FLAG-PIAS3 was kindly provided by Helene Boeuf and FLAG-Sp100 and EGFP-Sp100 by Hans Will.
Cell Culture and Transfections, Cell Lysis, GST Pulldown, SDS-PAGE, and Western Blot-All these techniques were done as described before (19). The HA antibody was used from Santa Cruz Biotechnology, FLAG M2 antibody form Sigma and GFP/YFP antibody form BD Bioscience.
Protein Expression and Purification for NMR Studies-Full-length SUMO1 and SUMO2 were cloned as GST fusions into pET-41a vectors (Novagen), expressed in bacteria on either LB medium (for non-marked protein) or minimal medium with 15 NH 4 Cl and [ 13 C 6 ]glucose for labeled protein. The recombinant protein was purified on a GSH resin (Amersham Biosciences) according to manufacturer's instruction and cleaved with thrombin, leaving the extension Gly-Ser in N-term of the full length SUMO; concentrators (Amicon) were used for tag removal and concentration. The protein was lyophilyzed and stored at Ϫ20°C.
NMR Spectra Acquisition and Assignment-All measurements were made at 27°C in 25 mM Phosphate buffer at pH 7. Triple resonance and two-dimensional experiments were performed on a Varian Inova 600 equipped with shielded Z gradients. Three-dimensional NMR spectra were processed using the standard Bruker software XWINNMR. Twodimensional NMR spectra were processed using NMRpipe (42). Analysis and visual representation of two-dimensional spectra were performed using Sparky (T. D. Goddard and D. G. Kneller, University of California, San Francisco), and three-dimensional spectra were analyzed with the program Aurelia (Bruker). Assignment of SUMO2 was generated using the spectra HNCA, HNCACB, CBCA(CO)NH, HC(CO)NH, and C(CO)NH.
NMR Titration Experiments-For peptide titration experiments heteronuclear single quantum coherence (HSQC) spectra were performed on 15 N-labeled SUMO1 or SUMO2 (300 mM in 25 mM phosphate buffer, pH 7). Unlabeled peptides were chemically synthetized by Thermo Electron GmbH, dissolved to a concentration of 2.7 mM in 25 mM phosphate buffer, pH 7, and titrated to the protein to reach a final peptide:SUMO ratio of 1.36:1. No precipitation could be observed even at the highest peptide concentrations. K D values were measured by two different methods, depending on the exchange regime. For amino acids in fast exchange regime, at each titration step, the distance of each peak from its original position was measured using the normalization pro-posed before (20). The obtained curves were fitted to a Hill 4-parameter model using the software SigmaPlot.
Bioinformatical Analysis-All sequence data base searches were performed with a non-redundant data set constructed from current releases of Swiss-Prot, TrEMBL, and GenPept (21,22). Generalized profile construction (23) and searches were run locally using the pftools package, version 2.1. Generalized profiles were constructed using the BLOSUM45 substitution matrix (24) and default penalties of 2.1 for gap opening and 0.2 for gap extension. The statistical significance of profile matches was derived from the analysis of the score distribution of a randomized data base (25).
Tryptic Digestion of PIAS and Detection of the Resulting Fragments by MALDI Spectrometry-HEK 293T cells were transfected with FLAG-PIAS and lysed, and the FLAG-PIAS expression was checked with Western blot analysis as described before. Immunoprecipitation of FLAG-PIAS was done with M2-agarose from Sigma according to manual instruction. Immunoprecipitated PIAS was loaded on SDS-PAGE gel and stained with Coomassie. The band containing PIAS was cut out of the gel and transferred into a microtube. As a negative control, a same sized bit of the gel cut from a region containing no protein and further handled as the probe. Except the digestion itself, all further steps were performed at room temperature under agitation (1000 rpm). The band was destained overnight using 0.5 ml of acetic acid/methanol/water 1:2:7 (v:v:v). It was washed for 4 h with water, dried with a SpeedVac, washed with (50% acetonitrile, 50% 0.2 M NH 4 HCO 3 in water, pH 8.9), and dried with a SpeedVac again. The band was soaked with 15 l of 0.2 M NH 4 HCO 3 in water, pH 8.9, containing 33 g⅐ml Ϫ1 trypsin proteomic grade (Sigma). The gel was reduced into little bits using the heat-rounded tip of a pasteur pipette and incubated at 37°C for 24 h. The bits were washed with water for 2 h, dried with a SpeedVac, and covered with 50 l of formic acid/water/isopropanol 1:3:2 (v:v:v) saturated with ␣-cyano-4-hydroxycinnamic acid. After overnight incubation, tubes were opened to allow crytallisation of the ␣-cyano-4-hydroxycinnamic acid together with the extracted peptides. The obtained solution was pipetted onto the MALDI plate avoiding to pipette gel bits. Measurements were made in reflector-positive mode, low mass gate set at 500 Da, and monitoring the 1-3-kDa range.

Identification of SUMO1-and SUMO2-interacting Partners-To
identify proteins that non-covalently bind to SUMO1 and SUMO2, we fused SUMO1 and SUMO2 mutants lacking two C-terminal glycine residues to the Gal4-DNA-binding domain of the YTH9 bait vector and performed large scale yeast two-hybrid screens using human thymus, spleen, and kidney libraries. After retransformation and X-gal tests we sequenced 102 SUMO1-interacting and 77 SUMO2-interacting partners. In this collection, we subsequently identified about 20 different candidate SUMO-interacting proteins, including SUMO-conjugating enzyme (UBC9), thymine DNA glycosylase (TDG), TOPORS, four members of the PIAS family, and RanBP2, which have been previously shown to interact with SUMO1 or SUMO2 (13,14,26). Altogether, we identified 10 new candidate SUMO-interacting proteins for which no data about sumoylation or SUMO interaction are available (Fig. 1A). Among these proteins there were several zinc finger-containing proteins including ZCCHC7, ZCCHC12, ZNF237, ZNF198, and ZHX1 involved in different processes like DNA repair or transcriptional repression (27,28). Moreover, Senataxin, a newly identified helicase mutated in patients suffering from ataxia-ocular apraxia 2, was found to interact with SUMO proteins in our screens (29). So far, two helicases, the BLM helicase and Werner's helicase, are known to be sumoylated and play important roles in causing hereditary diseases (30,31). The findings of new SUMO-interacting partners underline the important function of SUMO signals during transcription, DNA repair, and chromatin remodeling (32).
We next investigated whether these SUMO-interacting proteins bind preferentially to SUMO1 or SUMO2 in yeast cells. We retransformed all clones found to bind to SUMO1 in yeast expressing SUMO2 and in yeast expressing SUMO1 in parallel. The same was done for all the clones found to bind to SUMO2. As a control we transformed empty prey vector in yeast containing SUMO1 and SUMO2, while PIAS1, which was shown to interact strongly with SUMO1 and SUMO2, was transformed into the empty yeast strain. To compare the binding affinity to SUMO1 and SUMO2 we re-plated three different colonies of each retransformation on another agar plate with appropriate drop out medium. After 2 days yeast was transferred on a filter paper to compare the growth. Most of the retransformed SUMO-interacting partners bound to SUMO1 and SUMO2 with the same strength (Fig. 1B), whereas RanBP2 preferentially bound to SUMO1, an interaction that was already shown to be important for its function (33), and TTRAP (TRAF and TNF receptor-associated protein), strongly bound to SUMO2 and poorly to SUMO1 (Fig. 1B). Nothing grew on the control plates. To compare binding affinities of SUMO1 and SUMO2 with SUMO3, we did in vitro pulldown assays for PIAS1, PIASx␣, and UBC9 (Fig. 1C). As expected, SUMO3 bound with a similar affinity to these proteins as SUMO1 and SUMO2 did.
Characterization of SUMO-interacting Motif in TTRAP-TTRAP was originally found as a protein interacting with members of the tumor necrosis factor receptor (TNF-R) superfamily as well as several TRAFs (18). Since TTRAP bound more potently to SUMO2 than SUMO1 in yeast (Fig. 1B), we analyzed the interactions between SUMO1 and SUMO2 with TTRAP in more details. When cell lysates expressing HA-TTRAP were incubated with beads coupled to GST, GST-SUMO1, GST-SUMO2, or GST-SUMO3, TTRAP bound more potently to SUMO2 and SUMO3 than to SUMO1 (Fig. 2, A and B). TTRAP did not contain one of the published SUMO-binding domains. However, we could find the sequence I-V-D-V at positions 280 -284, which is the inversion of the proposed V/I-X-V/I-V/I FIGURE 1. Yeast two-hybrid screen. A, table of the newly identified SUMO-interacting proteins with their Swiss-Prot ID and name. B, growth of yeast containing SUMO1 or SUMO2; each prey plasmid was retransformed into yeast containing either SUMO1 or SUMO2 as bait in parallel. Yeast was grown on agar plates containing the appropriate drop out medium. After 4 days three different colonies were replated on another agar plate and incubated at 30 degrees. After 3 days yeast was transferred to a filter and growth was compared. First column, Swiss-Prot ID; second column, name of the prey; third column, three replicates of yeast clones, retransformed with prey and either SUMO1 or SUMO2. C, confirmation of yeast twohybrid results with GST pulldown assays for PIAS1, PIAS2 (isoform x␣), and UBC9 using HEK 293T cells transfected with PIAS1, PIAS2, or UBC9 and GST-SUMO1/2/3. GST alone was used as negative control. Western blots were performed with antibodies against the FLAG epitope and YFP. Levels of GST fusion proteins were determined by Ponceau S staining. TCL, total cell lysate.
SUMO-interacting motif (14). Recently, it was shown that this hydrophobic part could bind both parallel and antiparallel to SUMO so that these four amino acids could be a SUMO-binding domain as well (17). We mutated all four amino acids to alanine (HA-TTRAP-SIM*) and completely abolished binding to all SUMO isoforms (Fig.  2B). Since these binding assays were performed in yeast and mammalian cells, they raised a concern whether additional cellular proteins might contribute to indirect binding between SUMO and TTRAP. To verify their interaction in vitro system, full-length TTRAP and TTRAP-SIM* were expressed and purified as a GST fusion protein in bacteria and challenged with recombinant SUMO2. As shown in Fig. 2C, SUMO2 bound to GST-TTRAP but not to GST alone or GST-TTRAP-SIM*. This result confirmed that the SIM of TTRAP directly interacts with SUMO2 and that this signature motif is essential for SUMO binding to full size TTRAP.
Definition of a Universal SIM-Three different amino acid signature motifs have been proposed to mediate binding to SUMO (Fig. 3A). To identify minimal SIMs in newly cloned SUMO-interacting proteins, the sequences of the clones were subjected to bioinformatical analyses. All of the three proposed sequence motifs implicated in binding to SUMO are represented in both known and new SUMO-interacting partners pooled in our screens (Fig. 3B). Interestingly, several SUMO-interacting partners contain two SUMO-binding domains. Some SUMO-interacting partners, including TTRAP, MCAF, and ZCCHC12, do not contain the complete characteristics of any of the published domains but rather represent the inversion of motif 2 (Fig. 3B).
An alignment of all SUMO-interacting motifs revealed that they all harbored a hydrophobic core sequence consisting of stretches of three or four hydrophobic Ile, Leu, or Val residues plus one acidic/polar residue at position 2 or 3. The sequences surrounding this core-binding domain are predicted to be disordered and have a net negative charge due to a stretch of acidic amino acid residues (Fig. 3C). The stretch of acidic amino acids can be either at the C-or at the N-terminal site of the hydrophobic core. Furthermore, a spacer containing a conserved thre-onine can separate the hydrophobic part from the acidic one. Interestingly, the majority of SIMs contains one or more serines or threonines, being potential phosphorylation sites in vivo.
Biophysical Parameters Underlying Binding of SIMs to Different SUMO Paralogues-We used NMR spectroscopy to gain detailed insights into the binding of different SIM motifs to SUMO1 or SUMO2. In typical NMR titrations, binding of a ligand to a protein influences the environment of the atoms of the protein, especially in the binding interface between the protein and the ligand. Such perturbations are easily observable as modification of the frequency and intensity of resonances in HSQC spectra during a titration experiment. We used the already published assignment of SUMO1 (34) and measured a set of threedimensional spectra to assign the resonances of all atoms in the backbone and side chains of SUMO2 (supplemental Fig. 1).
To characterize the binding interface on SUMO paralogues and different SIMs we chose to study the SIMs of PIASx␣, which interact with equal strength with SUMO1 and SUMO2 (Fig. 1, B and C) and contain a stretch of acidic amino acids and several putative serine phosphorylation sites. We also analyzed the SIM of TTRAP, which interacts predominantly with SUMO2 in vivo and in vitro experiments and lacks the acidic tract. Thus, three different PIASx␣ peptides and one TTRAP peptide were synthesized to investigate the binding characteristics of SUMO1 and SUMO2 to these peptides ( Fig. 4A and supplemental  Fig. 2).
Most of the amino acids in slow exchange in the titrations of SUMO1 and SUMO2 with SIM PIASx␣ are found in the ␤ 2 -strand and ␣-helix of those proteins (Fig. 4, B and C). The K D associated with the amino acids in slow exchange is estimated to be 3 M for SUMO1 and 2 M for SUMO2, which is in good agreement with the results derived from isothermal titration calorimetry measurements (14).
The binding surface of the other peptides (SIM TTRAP , SIM pPIASx␣ , SIM PIASx␣ short ) was found to be on the same position on SUMO1 and SUMO2 showing that this surface represents a general binding surface for the SIM on SUMO (Fig. 4D). Since this surface is partly constituted by a side of the ␤-sheet of SUMO, and the hydrophobic core of the SIM has a sequence typical of a ␤-strand, we propose that the hydrophobic core of the SIM binds to the ␤ 2 -strand of SUMO by forming an intermolecular ␤-sheet.
Serine Phosphorylation in the SIM of PIASx␣ Regulates Its Binding to SUMO1-According to the K D values SIM PIASx␣ binds with similar affinity to SUMO1 and SUMO2. However, the curves of these two titra-tions have different shapes: Whereas the curves obtained in the titration of SUMO2 with SIM PIASx␣ have the expected shape for a simple 1:1 protein to peptide binding, the curves obtained in the titration of SUMO1 with this peptide have an unexpected sigmoidal shape (Fig.  5A). Such curves are also observed in the titration of SUMO1 with SIM PIASx␣ short . The titration curves of SUMO1 and SUMO2 with SIM pPIASx␣ have the classical shape, showing that the phosphorylation of the SIM is sufficient to change from one to the other binding  mode. Three explanations can be invoked for the sigmoid shape of the titration curves of SUMO1 with SIM PIASx␣ . The first possibility is that these peptides could have two binding interfaces on SUMO. However, our results show no evidence for a second binding site on SUMO, and the small size of SIM PIASx␣ short makes it unlikely that it binds SUMO through two different sites. The second possibility is that SIM PIASx␣ binds in different orientations to SUMO1, which can be ruled out according to the underlying calculations. Alternatively, those peptides have only one binding site on SUMO which conformation can be changed upon binding of the peptide (induced fit).
When SUMO1 titrations with SIM PIASx␣ and SIM pPIASx␣ are compared, it appears that the only amino acid that has a significantly different behavior in these two titrations is Lys 37 , which is in fast exchange with SIM PIASx␣ and in fast intermediate exchange with SIM pPIASx␣ . This shows that the phosphate group of SIM pPIASx␣ is likely to bind to SUMO1 in the neighborhood of this lysine (the same difference is observed for the Lys 34 of SUMO2, which is equivalent to the Lys 37 of SUMO1, confirming this hypothesis). The Lys 37 of SUMO1 and the Lys 34 of SUMO2 are situated at the end of the ␤ 2 -strand, which is much more bent in SUMO1 than in SUMO2. In close proximity to the Lys 37 of SUMO1 and the Lys 34 of SUMO1 is another lysine residue (Lys 39 in SUMO1 and Lys 36 in SUMO2) that cannot be observed due to HSQC peaks overlap and whose conformation is different in both SUMO isoforms: the Lys 39 side chain of SUMO1 points into the binding site, whereas the Lys 36 of SUMO2 points away from the SIM-binding site. Therefore it could be possible that the negatively charged phosphate group of SIM pPIASx␣ interacts with the positively charged Lys 39 of SUMO1 causing a conformational change favorable for SIM pPIASx␣ binding and strongly affecting the neighboring Lys 37 . The absence of a phosphate group in SIM PIASx␣ would make this transition more difficult to achieve. A high SIM PIASx␣ concentration would be required to maintain SUMO1 in the binding-favorable conformation, producing the observed two-step titration curve. Whether phosphorylation is present or not does not influence the binding of PIAS peptides to SUMO2 much, since the conformation of Lys 34 is already favorable. The recently published structures of SUMO1 in complex with a PIAS derived peptide (17) and with a RanBP2 derived peptide (16) confirm this hypothesis. In both structures the end of the ␤ 2 -strand of SUMO1 is less bent than in the free SUMO1 (6), and the Lys 39 side chain of SUMO1 moved away from the SIM-binding site to accommodate the peptide.
We note that the interaction between the Lys 37 of SUMO1 and the phosphate group of SIM pPIASx␣ indicates the orientation in which the SIM binds to SUMO: to take place while having the ␤ 2 -strand of SUMO binding the hydrophobic core of the SIM, those two elements must be parallel to each other. This has been recently verified by the structure published by Song et al. (17).
PIASxa Is Phosphorylated within the SIM in Vivo-These biophysical findings indicate that serine phosphorylation of the SIM may be relevant for in vivo interactions between SUMO1 and the SIM PIASx␣ . Accordingly, mutations of corresponding serines to alanines in the SIM PIASx␣ revealed that these serines are required for its binding to SUMO1 in yeast cells (13). We therefore investigated the phosphorylation pattern of the SIM of PIASx␣ in cells. MALDI fingerprinting was used to verify in vivo phosphorylation of PIASx␣ at the putative phosphorylation site within the SIM. Masses corresponding to different phosphorylated fragments upon trypsin digestion contained in the SIM were observed in several spectra, in addition to less frequently observed masses corresponding the same fragments without phosphate incorporated ( Fig. 5E; a more detailed table is given in the supplemental Fig. 3). This shows that the PIASx␣ proteins are phosphorylated within the SIM in vivo and indicates that this modification may be of functional importance for binding of SUMO1 to SIM in PIAS proteins.
We next tested whether phosphorylation of SIM is essential for binding to SUMO isoforms in GST pulldown assays. In contrast to previously published data whereby mutations of the corresponding serines to alanines in the SIM of PM-Scl75 blocked their binding to SUMO1 in yeast cells (13), we have not observed a significant decrease upon muta-tion of all three serine residues to alanine within the SIM motif of PIASx␣ (data not shown). This could be explained by a compensatory interaction between the negatively charged amino acid tract of SIM PIASx␣ and SIM pPIASx␣ with Lys 78 of SUMO1 (Figs. 5, B and C). This interaction results in the similar affinity of non-phosphorylated and phosphorylated PIAS peptides binding to SUMO1 and SUMO2 (see below). Therefore, it is possible that PIAS phosphorylation in vivo may modulate the spatial orientation rather then affinities of PIAS binding to its sumoylated targets.
Contribution of Acidic Amino Acids in SIMs for SUMO1 and SUMO2 Binding-A number of negatively charged amino acids (Glu, Asp) are present in the SIM of most proteins found in the yeast two-hybrid screens, which indicates that they may play a regulatory role in binding to SUMO. To investigate their role, we studied the binding of SUMO to SIM PIASx␣ short (Fig. 4A), a variant of the PIASx␣ peptide lacking the negatively charged amino acids tract. This peptide binds to SUMO1 with much lower affinity than SIM PIASx␣ (no resonance of any amino acid in the SIM-binding site is observed to be in the slow exchange regime) (Fig. 4B). However, the resonance of amino acids Ile 22 , Phe 36 , Val 38 , Lys 39 , Lys 45 , and Ser 50 are in intermediate exchange. The K D value can be estimated to be higher than for SIM PIASx␣ and SIM pPIASx␣ and lower than 0.2 mM, a value obtained from titration curves of amino acids of SUMO outside the SIM-binding site. When studying the titration of SUMO1 with SIM PIASx␣ and with SIM pPIASx␣ , the resonance of Lys 78 (situated in the loop between the ␤ 3 -and ␤ 4 -strands) was in slow exchange, whereas it was not in the SIM PIASx␣ short titration experiment. This further supports the notion of an interaction between the negatively charged amino acid tract of SIM PIASx␣ and SIM pPIASx␣ with Lys 78 of SUMO1 (Fig. 5, B and C). This interaction results in the above described higher affinity of SUMO1 for SIM PIASx␣ and SIM pPIASx␣ than for SIM PIASx␣ short . In contrast, results obtained in the titration of SUMO2 with SIM PIASx␣ short are fairly similar to those obtained in the titration of SUMO2 with SIM PIASx␣ and SIM pPIASx␣ . Furthermore, SIM TTRAP , which has no tract of negatively charged amino acids, binds better to SUMO2 than to SUMO1. Taken together, these data show that negatively charged residues in the SIM make an important contribution to binding to SUMO1 but are only little involved in SUMO2 binding.
Acidic Amino Acids Influence Binding to SUMO1 but Not to SUMO2 in Vitro and in Vivo-To examine whether the observation that acidic amino acids of SIMs influence binding to SUMO paralogues is a general phenomenon, we performed mutation studies and analyzed SUMO binding in both yeast and pulldown assays. We deleted the acidic part of several SUMO-interacting partners in a similar way as PIAS short peptide was created (Fig. 6A). These mutants were transformed into yeast or were used for GST pulldown assays. The yeast deletion mutants (Sp100, Senataxin, PIAS4, TOPORS) were transformed into yeast expressing SUMO1 and SUMO2 in parallel with the corresponding wild type SUMO-interacting partners. To compare the growth of yeast three different colonies of each retransformation were replated on a fresh agar plate with the same dropout medium. After 2 days the yeast was transferred to a filter paper, and the growth of yeast expressing SUMO1 or SUMO2 and either wild type or mutant constructs was compared (Fig.  6B). In all cases yeast containing SUMO1 and the acidic deletion mutants grew much slower than yeast containing SUMO2 and the deletion mutants, while yeast transformed with the wild type clones grew with equal speed in both cases. Furthermore we deleted acidic amino acids of the SIM in several SUMO-interacting partners (Sp100, PIAS1, PIAS3) and expressed them in mammalian cells. In GST pulldown assays with GST-SUMO1, GST-SUMO2, and GST-SUMO3, the results were similar to those obtained with the yeast two-hybrid system: bind-ing to GST-SUMO1 was reduced or even abolished, whereas binding to SUMO2/3 was unaltered (Fig. 6C). Taken together, these results show that acidic amino acids and negative charges are important for binding to SUMO1 but not to SUMO2/3.
The SIM of TTRAP does not contain a negative amino acid tract (Fig.  3C) and was shown to preferentially bind to SUMO2 compared with SUMO1 both in vitro and in vivo (Figs. 1B, 2A, and 4B). We were further interested in analyzing whether addition of acidic charged amino acids in TTRAP would switch its binding preference toward SUMO1 in the context of full protein. The amino acids Leu 286 , Gly 287 , and Lys 288 were mutated to the negatively charged acidic amino acids Glu, Asp, and Glu, respectively. In GST pulldown assays of TTRAP wild type and TTRAP mutant, we found that the negatively charged SIM of the TTRAP mutant gains ability to bind to SUMO1, whereas the binding to SUMO2 decreased compared with the binding between TTRAP wild type and SUMO2 (Fig. 6D). Thus, in this case introduction of negatively charged amino acids shifts the interaction from preferential SUMO2 to SUMO1 binding.

DISCUSSION
Protein modifications mediated by conjugation of SUMO to target proteins represent an emerging mechanism by which cells control distinct cellular functions (35). While our understanding of the mechanisms of protein conjugation by SUMO are quite advanced (4, 16, 33, 36), much remains to be understood on how these modifications are translated into different biological responses. It has already been described that SIMs bind to SUMO1 via their hydrophobic core. In this report we describe molecular and biophysical parameters underlying the interactions between SUMO1 and SUMO2 and SIMs in more detail.
Several recent publications nicely describe the binding of the hydrophobic core to be the main mediator of SUMO binding. Our mutation studies in TTRAP support these results indicating that few hydrophobic amino acids are sufficient to mediate binding to SUMO. However, our in vivo studies in yeast show a completely different picture: only 1 out of 20 different yeast clones exclusively contains the hydrophobic core. All the others additionally contain acidic amino acids and sometimes putative phosphorylation sites. This suggests that amino acids surrounding the hydrophobic core also influence binding to SUMO.
Three different SUMO isoforms are expressed in eukaryotes, which differ mostly in the amino acid composition of the ␤ 2 -strand and the ␣-helix (7), exactly the regions that we found to mediate binding to SIMs. This surface of SUMO can thus be regarded as a "code of specificity" of SUMO isoforms for the SUMO-SIM interaction. Interestingly, this region has been shown to be critical for the transcriptional inhibitory properties of SUMO (37). As the signature of the SIM motif is limited to the short ␤-strand forming motif, it should be possible to find the intermolecular strand either in parallel or antiparallel orientation, as long as backbone hydrogen binding can occur and the hydrophobic side chains are arranged in an inverse (reverse) manner. Indeed, we and others (16,17) found the PIASx␣ peptides bound parallel to the FIGURE 6. Deletion of the acidic amino acids of the SIM influences SUMO1 much stronger than SUMO2 binding. A, acidic amino acids were deleted in several SUMO-interacting partners according to the creation of the PIAS short peptide described before. The table shows the deleted part and amino acids. B, wild type and mutants of SUMO-interacting partners were retransformed in yeast containing SUMO1 and SUMO2 in parallel. Growth was analyzed as described before (Fig. 1B), C, lysates from HEK 293T cells expressing wild type FLAG-SP100, FLAG-PIAS1, or FLAG-PIAS3 and their corresponding mutants lacking the acidic amino acids of the SIM (EGFP-SP100 MUT, FLAG-PIAS1 MUT, or FLAG-PIAS3 MUT) were incubated with GST alone or GST-SUMO1/2/3 and analyzed by SDS-PAGE and subsequent Western blotting with indicated antibodies. Levels of GST fusion proteins were determined by Ponceau S staining. TCL, total cell lysate. D, GST pulldown of HA-TTRAP wild type and HA-TTRAP mutant containing additional acidic amino acids. The mutant with acidic amino acids binds stronger to SUMO1 but not to SUMO2, which is opposite from the wild type protein.
␤ 2 -strand, whereas the RanBP2 fragment was shown to bind antiparallel to SUMO. Since the hydrophobic core of the SIM is preceded by acidic residues in RanBP2 and followed by acidic residues in PIAS, it can be expected that the position of the acidic residues tract observed in SIMs determine their orientation relatively to the ␤ 2 -strand of SUMO. The negative charge borne by the acidic residues tract can be furthermore supported by phosphorylation of serine residues within the SIM. Phosphorylation, as well as negatively charged acidic residues, help to maintain specificity and orientation of binding of SUMO1 to different ligands thereby increasing the affinity by providing additional electrostatic interactions.
In contrast to SUMO1 the affinity of SUMO2 does not depend on the presence of an acidic track or phosphorylation and is mainly defined by its core SIM signature. However, also SUMO2 binding does not seem to be exclusively mediated by the hydrophobic core. Both RanBP2 and the TTRAP mutant containing additional acidic amino acids do not bind to SUMO2 indicating that also in the case of SUMO2 the hydrophobic core of the SIM is not exclusively responsible for mediating binding. It remains to be investigated which residues surrounding the hydrophobic core are crucial for SUMO2 binding.
The presented concept of surrounding charges determining the orientation of a peptide binding to SUMO resembles the binding mode of polyproline stretches when interacting with SH3 domains. The polyproline type II helices bind to SH3 domains in both N-C and C-N directions while forming similar hydrogen bond networks. In analogy to the basic amino acid tract flanking PXXP motifs, a negatively charged tract assists binding in the case of SIM motifs. The salt bridge formed between a positively charged residue in the flanking region of the polyproline type II helix to a negative residue in the SH3 domain resembles the strong binding between Lys 78 of SUMO1 and the negatively charged residues following the hydrophobic core of the SIM of PIASx␣. The similarity found in the binding of polyproline stretches and SH3 to the SIM-SUMO interaction entails us to define SIM ligands binding in antiparallel orientation to the ␤ 2 -strand of SUMO as class I ligands (RanBP2) and those binding in parallel orientation as class II ligands (PIASx␣).
Although the structures of SUMO and Ub are very similar, the binding surfaces of SUMO-SIM versus Ub-binding domains (UBDs) are clearly different (Fig. 7). At present, all currently tested SIMs show exclusive binding to SUMO isoforms but not to Ub or other Ub-like molecules. Although two other Ub-like proteins (NEDD8 and FAT10) have been implicated to bind to UBDs (38,39), no known Ub-binding domain was shown to associate with SUMO isoforms, thus implicating that SUMO-SIM recognition is much more specific than Ub/NEDD8/ FAT10 association with UBDs. In contrast to the formation of an intermolecular SIM-SUMO ␤-sheet, the interaction between UBDs and Ub is defined by a hydrophobic surface containing Ile 44 (40). This hydrophobic surface would correspond to the ␤ 3 -strand of SUMO, which is located to the opposite side of the molecule in relation to the ␤ 2 -strand where the SIM binds (Fig. 6). In principle all amino acids in the close vicinity of Ile 44 , which are exposed to the surface, can serve as key elements defining the specificity of UBDs. Despite these structural differences, the SUMO-SIM and the Ub-UBD interactions serve the same purpose: they are signaling pairs that transmit intracellular signals and regulate numerous cell functions.
In conclusion, this study describes the molecular details of specific interactions between SUMO1 and SUMO2 paralogues and SIMs and indicates the rational for their functional differences in vivo.