Biochemical Journal

Research article

Monitoring protein–protein interactions in mammalian cells by trans-SUMOylation

Ratnesh K. Srivastav , Susan Schwede , Malte Klaus , Jessica Schwermann , Matthias Gaestel , Rainer Niedenthal

Abstract

Protein–protein interactions are essential for almost all cellular processes, hence understanding these processes mainly depends on the identification and characterization of the relevant protein–protein interactions. In the present paper, we introduce the concept of TRS (trans-SUMOylation), a new method developed to identify and verify protein–protein interactions in mammalian cells in vivo. TRS utilizes Ubc9-fusion proteins that trans-SUMOylate co-expressed interacting proteins. Using TRS, we analysed interactions of 65 protein pairs co-expressed in HEK (human embryonic kidney)-293 cells. We identified seven new and confirmed 16 known protein interactions, which were determined via endogenous SUMOylation sites of the binding partners or by using SUMOylation-site tags respectively. Four of the new protein interactions were confirmed by GST (glutathione transferase) pull-down and the p38α–Edr2 interaction was verified by co-localization analysis. Functionally, this p38α–Edr2 interaction could possibly be involved in the recruitment of p38α to the polycomb chromatin-remodelling complex to phosphorylate Bmi1. We also used TRS to characterize protein-interaction domains of the protein kinase pairs p38α–MK2 [MK is MAPK (mitogen-activated protein kinase)-activated protein kinase] and ERK3 (extracellular-signal-regulated kinase 3)–MK5 and of the p38α–p53 complex. The ability of TRS to monitor protein interactions in mammalian cells in vivo at levels similar to endogenous expression makes it an excellent new tool that can help in defining the protein interactome of mammalian cells.

  • binding domain
  • in vivo analysis
  • in vivo labelling
  • protein–protein interaction
  • trans-SUMOylation (TRS)
  • Ubc9 fusion-directed SUMOylation

INTRODUCTION

Specific and regulated protein–protein interaction is essential for the existence of productive multi-protein complexes that maintain and regulate almost all processes in the cell, from sugar, fat and amino acid metabolism to transcription, translation, replication, chromosome segregation and cytokinesis. Our understanding of these processes mainly depends on the identification and characterization of the specific proteins involved and on the properties of their interactions. Various methods have therefore been developed which qualitatively and/or quantitatively analyse protein–protein interactions (reviewed in [1,2]). The most prominent methods, which have also been adapted for proteome-wide analysis, are the Y2H (yeast two-hybrid) system [3] and affinity-purification combined with MS (reviewed in [4]). There are several versions of both methods, and each has different advantages and limitations. The isolation of endogenous protein complexes using specific antibodies for co-immunopurification could give insights into protein complexes close to the physiological reality. However, appropriate antibodies are rare, and proteins that are weakly expressed or that have weak transient interactions are hardly detectable. These problems are partially solved by the use of transiently or stably overexpressed tagged proteins [58]. However, these may also increase the number of false-positive interactors, as some protein interactions only take place during the protein-extract preparation and therefore do not correspond to the actual conditions in the living cell.

Alternatively, Y2H represents a genetic approach that also identifies weak binary protein interactions and directly provides the DNA coding for the interacting proteins. It was developed for the yeast Saccharomyces cerevisiae, but was also adapted to Schizosaccharomyces pombe and to Escherichia coli. Y2H has been used for the identification of thousands of protein–protein interactions of a variety of organisms (summarized in [1]). However, in these approaches, proteins are expressed in a heterologous organism (yeast or E. coli), most of the native post-translational modifications and chaperone protein interactions are lacking and, in yeast, the system also depends on the transport of the interacting proteins to the nucleus. Accordingly, after nearly two decades of extensive use of Y2H, this method is now estimated to produce between 25 and 45% of false-positive results [9]. Further in vitro methods for monitoring protein–protein interactions, such as pull-down assays, peptide microarrays or protein co-crystallization, are relatively far from physiological reality. Other methods, such as fluorescence spectroscopy and imaging technologies, attempt to analyse the interaction of proteins in their native environment. New imaging technologies coupled with genetically constructed fluorescent proteins are used for direct co-localization of proteins by FRET (fluorescence resonance energy transfer) [10,11] or by PCAs (protein fragment complementation assays) with split-GFP (green fluorescent protein) [12]. These elegant methods open new opportunities to analyse single protein–protein interactions but, so far, they seem not to be suited for general or proteome-wide analysis (summarized in [13]).

In the present paper, we describe TRS (trans-SUMOylation), a new method that monitors protein–protein interactions in vivo and operates under protein expression levels similar to that of the endogenous proteins. In TRS, the fusion of the SUMOylating enzyme Ubc9 to the protein of interest (protein 1) results in covalent modification of an interacting protein (protein 2) by SUMOylation, a ubiquitin-like post-translational modification of eukaryotic cells where a mature SUMO (small ubiquitin-related modifier) protein of 101 amino acids (SUMO1) is reversibly connected to a target protein via a covalent isopeptide bond [14]. TRS is based on the recently described UFDS (Ubc9 fusion-directed SUMOylation) system [15,16] that shows strong cis-SUMOylation of substrate proteins fused to Ubc9 (Figure 1A). When analysing the influence of protein SUMOylation on protein–protein interactions, we were surprised to find that Ubc9-fusion proteins can also induce the SUMOylation of binding partners in trans (TRS) (Figure 1B). The analysis carried out is sufficiently detailed and universally applicable as to prove that TRS is an excellent tool for monitoring protein–protein interactions of co-expressed proteins in mammalian cells.

Figure 1 Schematic representation of the TRS system

(A) In the UFDS, a protein directly fused to Ubc9 is SUMOylated by the SUMO-loaded Ubc9 in cis. (B) A protein 2 that binds to protein 1 can be SUMOylated in trans by the SUMO-loaded Ubc9 fused to protein 1.

EXPERIMENTAL

Plasmids

The expression plasmids for HA (haemagglutinin)–Ubc9, STAT1 (signal transducer and activator of transcription 1)–Ubc9, p53 and p53–Ubc9 were as described by Jakobs et al. [15], Ubc9–CDK4 (cyclin-dependent kinase 4), CRSP9 (cofactor required for Sp1 transcriptional activation, subunit 9)–Ubc9, CSNK2B (casein kinase 2, β polypeptide)–Ubc9, Fos–Ubc9, GST (glutathione transferase)–CSNK2B, GST–CDC37 (cell division cycle 37), GST–CDK4, GST–CRSP9, GST–DRG1 (developmentally regulated GTP-binding protein 1), GST–Fos, GST–HSF2BP (heat-shock factor 2-binding protein), GST–TAF10 [TBP (TATA-box-binding protein)-associated factor 10], GST–PSMC3 (proteasome 26S ATPase subunit 3-interacting protein), MEKK1ca {constitutively active MEK [MAPK (mitogen-activated protein kinase)/ERK (extracellular-signal-regulated kinase) kinase] kinase 1}, Ubc9–CDC37, Ubc9–HSF2BP, Ubc9–TBP, Ubc9–TAF10, Ubc9–PSMC3 and GST–TAF10 were as described by Jakobs et al. [16], HA–MK5 (MK is MAPK-activated protein kinase) and GST–ERK3 were as described by Schumacher et al. [17]. STAT1–FKBP (FK506-binding protein) was as described by Zimnik et al. [18]. The expression plasmids FLAG–p38α [19], GST–p38α [20], Myc–MK2 [21] and HA–Hsp27 (Hsp is heat-shock protein) [22] have been described previously. The expression plasmids for ERK1, HA–ERK2, HA–Jun, HA–JNK2 (JNK is c-Jun N-terminal kinase) and HA–JNK1 were provided by Dr Kirsten Mielke (Hannover Medical School). The expression vectors Ubc9–p38α, Ubc9–MEK2, Ubc9–MEK1, Ubc9–ERK2, MEK1–Ubc9, MEK2–Ubc9, ERK2–Ubc9, GST–p53, GST–ERK2, HA–CSNK2B, Jun–Ubc9, GST–Edr2 and GST–MK2 were generated by way of cloning by recombination (Gateway® system), the expression plasmids for Ubc9–MK5, Ubc9–MK2, Ubc9–HDGF (hepatoma-derived growth factor), MK2–ST-(377–393) (ST is SUMOylation-site tag), MK2–ST-(377–391), MK2–Ubc9, p53K386R, p53K386R-3G–ST-(377–393), p53K386R-6G–ST-(377–393), p53K386R–ST(RAGP1), p53K386R–ST-(377–393), p53K386R–ST-(377–389), p53K386R–ST-(377–391) and UC (uncleavable) SUMO3 (SUMO3Q89P) by PCR amplification.

Transfection, cell lysis and Western blotting

HEK (human embryonic kidney)-293 cells were grown in DMEM (Dulbecco's modified Eagle's medium) supplemented with fetal bovine serum (10%), L-glutamine (2 mM), penicillin (100 units/ml) and streptomycin (100 μg/ml). Transfection of 50–80% confluent HEK-293 cells was performed in 12-well plates using polyethyleneimine (1 mg/ml) solution, using the method of Niedenthal [23]. Then, 800 ng of total transfection-purity expression vector DNA was diluted in 45 μl of serum-free DMEM. Unless stated otherwise, we usually used 200 ng of expression plasmid for the interacting proteins and for EGFP (enhanced GFP)–SUMO1 and 100 ng of the expression plasmid for Ubc9 and of the empty vector pcDNA3, to reach the DNA concentration for the optimal transfection mixture. A 3 μl volume of the vortex-mixed polyethyleneimine stock solution was added to the diluted DNA and mixed. After 10 min of incubation at room temperature (23 °C), 450 μl of fresh DMEM was added to the mixture and vortex-mixed. After removing the medium covering the cells, the transfection mixture was added, and the cells were incubated for 24 h at 37 °C at 5% CO2. After washing the cells with 1 ml of PBS at room temperature, the cells were lysed in modified Laemmli (2-fold) buffer [160 mM Tris/HCl (pH 6.8), 4% (w/v) SDS, 20% (v/v) glycerol, 0.5% 2-mercaptoethanol and 0.008% Bromophenol Blue). The samples then were heated for 10 min at 96 °C in a heat block. For Western blot analysis, the proteins were separated by SDS/PAGE (7.5–16% gels), then blotted on to a PVDF membrane and developed with a specific primary antibody, an HRP (horseradish peroxidase)-conjugated secondary antibody, the Immobilon® Western HRP substrate (Millipore) and the LAS-3000 imaging system (Fuji).

Protein kinase assay

Samples of 6 μg of the purified recombinant fusion protein GST–Bmi1 expressed in E. coli using pDEST15-Bmi1 or the appropriate GST–Bmi1 mutant were incubated with 1.8 μg of active recombinant GST–p38α in the absence or presence of 0.15 μg of His6–MK2. As a positive control, 10 μg of recombinant Hsp25 were incubated with 0.15 μg of His6-MK2 and 1.8 μg of active recombinant GST–p38α also in a reaction mixture of a final volume of 50 μl, containing 50 mM 2-glycerophosphate (pH 7.4), 0.1 mM EDTA, 10 mM magnesium acetate, 0.1 mM ATP and 1.5 μCi of [γ-33P]ATP. After 30 min at 30 °C, reactions were terminated by adding 16 μl of 4× SDS sample buffer. Labelling of proteins was analysed by SDS/PAGE (7.5–16% gels) and subsequent phosphoimaging using a Fuji Bas-1500.

Fluorescence microscopy

For subcellular localization of GFP-tagged protein or RFP (red fluorescent protein)-tagged proteins, the transfected cells were replated in Chambered Coverglass (Lab-Tek, Nunc) and analysed using a Leica DM IRBE microscope with the Leica TCS confocal systems program.

RESULTS

TRS monitors protein–protein interactions

SUMOylation substrate proteins are strongly SUMOylated when they are fused to the SUMO conjugating enzyme Ubc9 (UFDS) [15,16] (Figure 1A) and this protein fusion functionally can be replaced by a rapamycin-induced FRB (FKBP12–rapamycin-binding)/FKBP12 heterodimerization [18]. In the present study, we have analysed whether a Ubc9-fusion protein can also SUMOylate its binding proteins (TRS) (Figure 1B), thus making SUMOylation a monitor for protein–protein interactions in vivo. Therefore we first co-expressed a protein of interest (protein 1) fused to Ubc9 with an interacting protein (protein 2) in HEK-293 cells, which also express EGFP–SUMO1. As a result of interactions between protein 1 and 2 and subject to the condition that a SUMOylation site is available in protein 2, EGFP-SUMOylation will normally result in altered electrophoretic mobility of protein 2 and can be detected by Western blot analysis. To analyse TRS for its ability to identify protein interactions in mammalian cells in vivo, we tested proteinbinding pairs known from the literature [summarized in the BioGRID database (http://www.thebiogrid.org)] and Entrez Gene (http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=gene); for references, see Supplementary Tables S1–S4 at http://www.BiochemJ.org/bj/438/bj4380495add.htm)]. We always co-expressed protein 2 and a protein 1–Ubc9 fusion, or, as a negative control, non-fused and non-binding protein-fused Ubc9 in the absence and presence of EGFP-tagged SUMO1. Proteins were analysed for SUMOylation by Western blotting using protein 2-specific antibodies. Western blots demonstrating protein interactions detected by TRS are shown in Figure 2 and in Supplementary Figure S1 at http://www.BiochemJ.org/bj/438/bj4380495add.htm. In many cases, intense SUMOylation of the interacting proteins can be detected, indicating that TRS occurs as a general phenomenon (summarized in Supplementary Tables S1 and S2). It should be noted for all Western blots shown that the shift of a protein band by a SUMOylation is not simply as it is estimated by the addition of the mass of the conjugated SUMO or EGFP–SUMO1. Additionally, the localization of the conjugation site at the C/N-terminus or in the centre of the conjugated polypeptide determines how strong the protein band is shifted. Although conjugation to the ends of the polypeptide chain does not shift significantly more than due to the molecular mass added, conjugation to the centre of the polypeptide chain often leads to a significant extra shift. Using TRS, we confirmed 16 established protein interactions (Supplementary Table S1), such as interactions between the protein kinases MK2 and p38α (Figures 2A and 2B), between the tumour suppressor p53 and p38α (Figure 2C), the homodimerization of proteasome regulatory subunit PSMC3/PSMC3 (Figure 2D), between MK2 and the polycomb member Edr2 (Supplementary Figure S1A), between the AP-1 (activator protein 1) transcription factors c-Jun and c-Fos (Supplementary Figures S1B and S1C) and also between p53 and the regulatory kinase subunit CSNK2B (Supplementary Figure S1D and Supplementary Table S1). Furthermore, we identified seven new protein interactions (Supplementary Table S2): the homodimerization of HSF2BP (Figure 3A), of the regulatory kinase subunit CSNK2B (Figure 3B), of the Hsp90 co-chaperone CDC37 and of the protein kinase MK2 (Supplementary Figures S1E and S1F) as well as the interactions between MK2 and CSNK2B (Supplementary Figure S1G), MK2 and MK5 (Supplementary Figure S1G), and MK2 and ERK1 (Supplementary Figure S1H). Four of the new protein interactions identified, the homodimerization of HSF2BP (Figure 3C) and of CSNK2B (Figure 3D) as well as the interactions between MK2 and MK5 (Figure 3E) and MK2 and CSNK2B (Figure 3F), were confirmed by GST pull-down.

Figure 2 TRS identifies protein–protein interactions

(A) GST–MK2 and EGFP–SUMO1 or (B) GST–p38α and EGFP–SUMO1 were co-transfected into HEK-293 cells either alone (−) or with HA–Ubc9, p53–Ubc9, Ubc9–MK5, Ubc9–MK2, MK2–Ubc9, Ubc9–p38α, Jun–Ubc9 or Ubc9–CSNK2B (+). (C) p53 and EGFP–SUMO1 were co-transfected into HEK-293 cells either alone (−) or with HA–Ubc9, STAT1–Ubc9, ERK2–Ubc9, Ubc9–ERK2, Ubc9–TBP, Ubc9–CDK4 or Ubc9–p38α (+). (D) GST–PSMC3 and EGFP–SUMO1 were co-transfected into HEK-293 cells either alone (−) or with HA–Ubc9, Ubc9–PSMC3 or MK2–Ubc9 (+). At 24 h after transfection, the Ubc9-fusion protein-binding proteins in the protein extracts of the transfectants were detected by immunoblotting using protein or tag-specific antibodies (GST, p53 or Ubc9). After stripping, the Ubc9-fusion proteins were detected with anti-Ubc9 antibody (Ubc9). Binding proteins, binding proteins conjugated with co-expressed EGFP–SUMO1 (E-S1), Ubc9-fusion proteins and Ubc9-fusion proteins conjugated with co-expressed EGFP–SUMO1 (E-S1) are indicated by black arrowheads or black lines, weak TRS bands are indicated by grey open arrowheads. For all of the Western blots, it should be noted that the shift of a protein band by a SUMOylation is not simply as it is estimated by the addition of the mass of the conjugated SUMO or EGFP–SUMO1. Additionally, the localization of the conjugation site at the C/N-terminus or in the centre of the conjugated polypeptide determines strongly how much the protein band is shifted. The conjugation to the ends of the polypeptide chain does not shift strongly additional to the added mass, unlike the conjugation to the centre of the polypeptide chain. Molecular masses are indicated in kDa. G, GST; U, Ubc9.

Figure 3 Confirmation of protein–protein interactions identified by TRS

(A) GST–HSF2BP and EGFP–SUMO1 were co-transfected into HEK-293 cells either alone (−) or with HA–Ubc9, Ubc9–HDGF or Ubc9–HSF2BP (+). (B) GST–CSNK2B with (+) or without (−) EGFP–SUMO1 was co-transfected into HEK-293 cells either alone (−) or with CSNK2B–Ubc9 (+) and with (+) or without (−) MEKK1ca. At 24 h after transfection, the Ubc9-fusion protein-binding proteins in the protein extracts of the transfectants were detected by immunoblotting using a GST-specific antibody (GST). Binding proteins and binding proteins conjugated with co-expressed EGFP–SUMO1 (E-S1) are indicated by black arrowheads. (C) Ubc9–HSF2BP was co-transfected into HEK-293 cells either with GST or with GST–HSF2BP. (D) CSNK2B–FKBP was co-transfected into HEK-293 cells either with GST or with GST–CSNK2B. (E) GFP–MK5 was co-transfected into HEK-293 cells either with GST or with GST–MK2. (F) Ubc9–CSNK2B was co-transfected into HEK-293 cells either with GST or with GST–MK2. At 24 h after transfection, the GST and GST-fusion proteins bound to glutathione–Sepharose were pulled down from the protein extracts of the transfectants. Co-purification of Ubc9-, FKBP- and GFP-fusion proteins were detected by immunoblotting using tag-specific antibodies (Ubc9, FKBP and GFP). After stripping, the GST-fusion proteins were detected with anti-GST antibody (GST). The detected co-expressed proteins are indicated by black arrowheads. A non-specific band in (B) is indicated by a grey star. (G) Co-localization of p38α with Edr2. HEK-293 cells were transfected with expression vectors for GFP–Edr2 and RFP–p38α. Fusion proteins were visualized by fluorescence microscopy. Nuclei were stained using TO-PRO. RFP–p38α–AGF is the fusion protein of the catalytically inactive mutant of p38α which lacks the phosphorylatable threonine and tyrosine residues in the activation loop. p38α phosphorylates Bmi1 at Thr275. (H) In vitro phosphorylation assay for Bmi1 and Hsp25 using His6–MK2 alone and with its activating MAPK GST–p38α. Kinase assay was performed in the presence of [γ-33P]ATP and phosphorylated proteins were visualized by phosphorimaging. (I) Different potential phosphorylation site mutants corresponding to alanine replacement of the proline-directed serine/threonine residues shown in bold in the C-terminal sequence part of Bmi1 were subjected to in vitro phosphorylation using GST–p38α. Phosphorylation was detected as in (H). Molecular masses are indicated in kDa. WT, wild-type.

We have shown recently that MK2 associates with the PcG (polycomb group complex) and regulates via its interaction with Edr2 the haemopoietic stem cell homoeostasis [24]. When we verified MK2–Edr2 interactions by TRS and used p38α and Edr2 as control, we also become aware of a positive interaction signal for p38α and Edr2 in TRS (Supplementary Figure S1A). This novel interaction between p38α and Edr2 could be confirmed by nuclear co-localization of both proteins, when fused to GFP or RFP respectively and visualized using confocal microscopy (Figure 3G). Since Edr2 is part of the multi-protein PcG, we then analysed whether proteins in this complex are substrates of MK2 or p38α. Interestingly, p38α, but not MK2, phosphorylates the Edr2-interacting PcG protein Bmi1 at the proline-directed Thr275 (Figures 3H and 3I). Binding of p38α to Edr2 and most likely also to MK2 leads to its recruitment to the PcG complex where it can phosphorylate the PcG protein Bmi1. The Edr2-dependent docking of p38α to PcG and the subsequent phosphorylation of the complex component Bmi1 could indicate functional consequences and biological relevance of this novel interaction detected by TRS.

However, using EGFP–SUMO1 for 11 protein interactions described in the literature, such as STAT1–STAT1- (without Tyr701 phosphorylation), p53–JNK1/2- and the c-JUN–JNK2-interactions, TRS could not be detected (Supplementary Table S3 and Supplementary Figure S1).

TRS using a SUMOylation-site tag

There are several possible explanations for why some protein interactions could not be confirmed by TRS: first, TRS is not sensitive enough to detect very weak protein interactions, such as those found between enzymes and their substrates; secondly, the interactions identified under artificial conditions do not exist in vivo; thirdly, because of sterical restrictions and structural constraints in reaching the SUMOylation site by Ubc9; and, lastly, because there is no SUMOylation site present in the protein 2. To apply TRS for the latter two cases, we introduced an ST, which can introduce the ideal SUMOylation site of p53 into any protein of interest (Figure 4A). The ST-(377–393)–peptide TSRHKKLMFKTEGPDSD was added to the C-terminus of the non-SUMOylated mutant p53K386R (Figures 4A and 4B) with or without three and six glycine residues as linker respectively (Figure 4A). As expected, the ST-fusion leads to significant increased TRS of p53K386R-ST-(377–393) by Ubc9–p38α (Figure 4C and Supplementary Figures S2A and S2D at http://www.BiochemJ.org/bj/438/bj4380495add.htm) and by CSNK2B–Ubc9 (Supplementary Figure S2D) and of MK2–ST-(377–393) and MK2–ST-(377–391) by Ubc9–p38α (Supplementary Figures S2B and S2C). Furthermore, we tested other STs [ST-(377–389), ST-(377–391), ST(RAGP1) and ST8; Figure 4C and Supplementary Figure S2E) and demonstrated that even a peptide of eight amino acids (ST8) functions in TRS (Figure 4C). These data indicate that STs can be utilized to extend TRS analysis to proteins which do not contain endogenous SUMOylation sites.

Figure 4 TRS of SUMOylation-site tags

(A) Structure of p53 and the p53K386R mutant fused to the STs indicated. (B) p53 and EGFP–SUMO1 or p53K386R and EGFP–SUMO1 were co-transfected into HEK-293 cells either alone (−) or with Ubc9–p38α, Ubc9–BACH1 (BTB and CNC homology 1) or MK2–Ubc9 (+). (C) Ubc9–p38α and EGFP–SUMO1 were co-transfected into HEK-293 cells with p53wt (wild-type p53), p53K386R, p53K386R–ST-(377–393), p53K386R-3G–ST-(377–393), p53K386R-6G–ST-(377–393), p53K386R–ST-(377–391), p53K386R–ST-(377–389), p53K386R–ST(RAGP1) or p53K386R–ST8 (+). At 24 h after transfection, the Ubc9-fusion protein-binding proteins in the protein extracts of the transfectants were detected by immunoblotting using a p53-specific antibody. After stripping, the Ubc9-fusion proteins were detected with an anti-Ubc9 antibody (Ubc9). Binding proteins, binding proteins conjugated with co-expressed EGFP–SUMO1 (E-S1), Ubc9-fusion proteins and Ubc9-fusion proteins conjugated with co-expressed EGFP–SUMO1 (E-S1) are indicated by black arrowheads. A non-specific band in (C) is indicated by a grey star. Molecular masses are given in kDa.

TRS identifies interaction domains

Next we asked whether TRS can be used to characterize protein interaction domains. To that end, we analysed the specific interactions between the protein kinase MK5 and the atypical MAPK ERK3 (Figure 5), which was already characterized by us in pull-down experiments [17], also in TRS. First, we co-expressed Ubc9–MK5 and, as negative controls, also Ubc9–MK2 and Ubc9–p38α with full-length GST–ERK3 and analysed the protein extracts by Western blotting (Figure 5B). The known specificity of interactions between ERK3 and MK5 [17,25] was confirmed by TRS of GST–ERK3 detected for co-expression of Ubc9–MK5, whereas co-expression of Ubc9–MK2 and the Ubc9–p38α does not lead to TRS of GST–ERK3 in EGFP–SUMO1-expressing HEK-293 cells. Using pull-down experiments of MK5 with C-terminal ERK3-deletion mutants, it has been shown previously that the overlapping ERK3 regions between amino acids 301 and 357 [17] and amino acids 330 and 340 [25] are essential for MK5 binding. In the present study, we analysed TRS of the C-terminal ERK3-deletion mutants, as shown in Figure 5(A), and obtained confirmatory results with this new method: whereas Ubc9–MK5 leads to TRS of the ERK3-(1–471) and ERK3-(1–357) fragments (Figure 5B), no TRS of GST–ERK3-(1–301) is observed (Figure 5B). In addition, we were able to show that there is no TRS of the entire C-terminus of ERK3-(317–720) and of ERK3-(101–720) (Figure 5B) by Ubc9–MK5, indicating that binding of MK5 to ERK3 requires both parts of the kinase domain of ERK3 and parts of its C-terminal extension. The analyses of MK2–p38α and p53–p38α interactions by TRS also confirmed the known molecular facts (Supplementary Figure S3 at http://www.BiochemJ.org/bj/438/bj4380495add.htm)

Figure 5 Protein–protein interaction domains characterized by TRS

(A) Schematic representation of ERK3-deletion mutants fused to GST. (B) TRS analysis of ERK3-deletion mutants fused to GST. GST–ERK3 and EGFP–SUMO1 were co-transfected into HEK-293 cells either alone (−) or with HA–Ubc9, Ubc9–p38α, Ubc9–MK2 or Ubc9–MK5 (+). GST–ERK3-(101–720), GST–ERK3-(1–301), GST–ERK3-(316–720), GST–ERK3-(1–471) or GST–ERK3-(1–357) were co-transfected into HEK-293 cells with HA–Ubc9 and EGFP–SUMO1 or Ubc9–MK5 and EGFP–SUMO1 (+). At 24 h after transfection, the Ubc9-fusion protein-binding proteins in protein extracts of the transfectants were detected by immunoblotting using a GST-specific antibody (GST). After stripping, the Ubc9-fusion proteins were detected with an anti-Ubc9 antibody (Ubc9). Binding proteins, binding proteins conjugated with co-expressed EGFP–SUMO1 (E-S1), Ubc9-fusion proteins and Ubc9-fusion proteins conjugated with co-expressed EGFP-SUMO1 (E-S1) are indicated by black arrowheads or black lines. Molecular masses are indicated in kDa. G, GST.

TRS of proteins expressed at endogenous levels

Protein–protein interaction analyses are prone to false-positive results, particularly at the time when proteins are mixed in high concentrations or when they are co-overexpressed. To show that TRS can determine protein interactions under native in vivo conditions at expression levels that are similar to the endogenous ones, we analysed three different TRS pairs for protein–protein interactions at protein levels similar to endogenous ones. Therefore the indicated amounts (330 pg–10 ng) of specific fusion protein expression vectors were co-transfected with the indicated amounts of co-expression plasmids into HEK-293 cells (Figure 6). Under these conditions and using UCSUMO3 that reduces deconjugation, we reinvestigated the TRS of GST–MK2 by p38α–Ubc9 (Figure 6A), of GST–p38α by MK2–Ubc9 (Figure 6B) and of GST–ERK3 by MK5–Ubc9 (Figure 6C). When the GST- and Ubc9-fused proteins were expressed at levels similar to the non-fused endogenous proteins, we detected a weak TRS band (marked by open grey arrowheads) in all lanes where the GST-fusion protein was co-expressed with the Ubc9-fusion protein and SUMO (EGFP–SUMO1 and UCSUMO3). No SUMOylation was detected with the massive overexpression of Ubc9 alone (controls of Figures 6A and 6B), demonstrating the specificity of TRS. This proves that TRS can determine real in vivo interactions of proteins at expression levels of endogenous proteins.

Figure 6 TRS of proteins expressed similar to endogenous levels

(A) GST–MK2 and UCSUMO3, (B) GST–p38α and EGFP–SUMO1 or UCSUMO3 and (C) GST–ERK3 and EGFP–SUMO1 or UCSUMO3 were co-transfected into HEK-293 cells either alone (−) or with the indicated amounts of expression plasmids of HA–Ubc9, Ubc9–p38α, MK2–Ubc9 or Ubc9–MK5 (+) to reach expression levels similar to those of endogenous proteins. At 24 h after transfection, the Ubc9-fusion protein-binding proteins and the endogenous proteins in the protein extracts of the transfectants were separated by SDS/PAGE (7.5% gels) and detected by immunoblotting using protein-specific antibodies (MK2, p38α, Ubc9, ERK3 and MK5). After stripping, the Ubc9-fusion proteins and the endogenous (endog.) proteins were detected with specific antibodies (MK2, MK5, p38α and Ubc9). The Ubc9 and the Ubc9-fusion proteins were also detected after a separation by SDS/PAGE (13% gels) by immunoblotting using the anti-Ubc9 antibody (Ubc9). Binding proteins, binding proteins conjugated with co-expressed EGFP–SUMO1 (E-S1) or UCSUMO3 (S3), Ubc9-fusion proteins, HA–Ubc9, endogenous Ubc9, MK2, MK5, ERK3 and p38α are indicated by black arrowheads. UCSUMO3 represents the human SUMO3Q89P mutant. Weak TRS bands are indicated by grey open arrowheads. Non-specific bands in (A) and (B) are indicated by grey stars. Molecular masses are indicated in kDa. G, GST.

DISCUSSION

The identification of protein binding partners is an essential step towards the functional characterization of a protein. In the present paper, we describe TRS, a new method to identify and study in vivo protein–protein interactions by co-expression in mammalian cells. TRS is based on the finding that Ubc9 fusion proteins can trans-SUMOylate their binding partners. For 16 known and seven new protein interactions, we have demonstrated that TRS can be used to monitor the binding of the proteins in vivo and that TRS can also contribute to the characterization of in vivo interacting regions of protein pairs, such as MK5–ERK3, MK2–p38α and p53–p38α. Although the analyses of the MK5–ERK3 complex completely confirmed the known interacting domains, we cannot exclude that the used deletions in ERK3 possibly removed the SUMOylation site(s) in that protein. This possibility was excluded in the analyses of the MK2–p38α and p53–p38α complex, where the deletions of protein 1 have been fused to Ubc9 and the SUMOylation substrate protein 2 remains untouched. For future analyses of interaction domains by TRS, we would always suggest to tag protein 2 with a SUMOylation site to preserve its SUMOylatability.

Of note, using the protein pairs MK2–p38α and MK5–ERK3, which were moderately expressed in HEK-293 cells, we have demonstrated that TRS can detect the interactions of proteins at levels similar to their endogenous expression. Hence TRS could be regarded as an in vivo labelling mechanism for co-expressed proteins that interact with a specific Ubc9-fusion protein under most native conditions within the cell. The analysis that we have shown in the present study for moderately expressed proteins p38α, MK2, MK5 and ERK3 could also be adapted to lower expressed proteins, if the protein 2 that is analysed for SUMOylation is enriched before the Western blot analysis using a suitable tag.

What are the advantages and disadvantages of TRS compared with other methods for studying protein–protein interactions in vivo in the living cell? Often, it is suggested that co-immunoprecipitation of endogenous interacting proteins is the most suited system to confirm protein–protein interactions. Although this method starts with the native non-manipulated cellular system, immunoprecipitation of endogenous interacting proteins is performed after cell lysis in protein extracts, where proteins, which are normally separated by their distinct cellular localization or by local variations of physicochemical conditions, can artificially interact. Similar limitations exist for the methods that show protein–protein interactions by purification of protein complexes under native conditions, such as pull-down, TAP (tandem affinity purification) [26], the Cross-and-Capture method [27] and the Lumier system [28], and also the QUICK system [29], a method that can minimize non-specific contaminants, but cannot reduce false-positive interactions that exclusively take place in the protein extract. TRS minimizes the probability of such false-positive results, because SUMO-labelling takes place in an interaction-dependent manner only in living cells and because the conjugating enzyme Ubc9 is directly inactivated during cell lysis by boiling in SDS loading buffer. As a further control, we replaced the Ubc9-fusion protein by an overexpressed unfused Ubc9. This enabled us to clearly differentiate between Ubc9-dependent and Ubc9-fusion protein-dependent SUMOylation of the target protein.

Of the genetic methods for analysis of protein–protein interactions, the most relevant are the Y2H approaches. These analyse protein–protein interactions in the living yeast cell, where various modifications of mammalian proteins necessary for the interactions cannot be generated and where yeast proteins can possibly interfere with the mammalian protein–protein interactions analysed. Therefore mammalian two-hybrid systems have been developed, which are based on the reconstitution of a transcription factor [30], of the signalling of a membrane-spanning receptor signalling cascade [MAPPIT (Mammalian Protein–Protein Interaction Trap)] [31], by the protein fragment complementation of an enzymatic activity [split TEV (tobacco etch virus)] [32], by the BIFC (bimolecular complementation of a fluorescence signal) [33] or also by FRET/BRET (bioluminescence resonance energy transfer) [34,35]. These methods may all better reflect the real situation in the cell and have been used successfully to identify interacting proteins. However, their limitations arise from the facts that interactions have to take place in a specific cellular localization (mammalian two-hybrid, MAPPIT, split TEV and Tango) and that there are structural constraints to reach the protein fragment complementation (split TEV, BIFC and FRET/BRET). Furthermore, most of these methods are necessarily combined with an indirect reporter system that can also cause false-positive results. TRS is not restricted to a special cellular localization for the protein–protein interactions and it does not need a special reporter system, because the resulting modification is detected directly at the interacting protein.

Eleven protein interactions described previously could not be confirmed by TRS using EGFP–SUMO1. Possible reasons for this could be the existence of structural constraints between the binding proteins, a problem that concerns all systems based on fusion proteins. Our observation that TRS is not efficient with all constructs tested for the p53–p38α or the p53–ERK2 binding pairs (Supplementary Table S1) supports the notion of sterical constraints of TRS. We tried to minimize these sterical constraints for TRS by the development of special linkers that will give the fused Ubc9 more of the flexibility necessary to reach the SUMOylation site of the associating proteins but so far without success (results not shown). However, we could demonstrate that both sterical constraints and missing SUMOylation sites in the protein 2 can be compensated for by the introduction of highly flexible STs into protein 2. STs enable TRS of proteins without natural SUMOylation site(s), as shown for p53K386R, and strongly enhance the TRS of other proteins 2, such as MK2. In this context, it also should be noted that the Ubc9-fusion leads to a forced SUMOylation of the fusion partner in UFDS [15,16] and also to the interaction partner in TRS (the present study). Hence even weak SUMOylation sites can display TRS. However, a further explanation for missing protein 2 SUMOylation could be that some transient interactions between enzymes and substrates, such as JNKs–Jun, MK2–Hsp27 or ERK1–p53 (Supplementary Table S2), are too weak to be identified by TRS using EGFP–SUMO1. Further developments of TRS are under way, using SUMO with a reduced deconjugation, STs with an increased SUMOylation or a linker-fused Ubc9 with higher conjugation activity.

TRS in its present form depends on the co-expression of the interacting proteins, where the prey protein should contain a SUMO site or tag, and is surely suited to confirm protein interactions in vivo that have been identified with systems such as Y2H. Furthermore it can be used for a systematic co-expression of protein pairs to identify new interacting protein pairs in vivo. To ensure that as much as possible interactions will be identified, we suggest to use N- and C-terminal fusions of Ubc9 to a certain protein 1 and a protein 2 that carries an ST. TRS determines the protein–protein interactions of interest in the homologue organism and rules out misleading results caused by the heterologous organism used.

TRS is a completely new concept that uses a mammalian post-translational modification system to covalently label co-expressed interacting proteins in vivo and is capable of studying protein–protein interactions at levels similar to endogenous expression. Therefore TRS has the potential to contribute significantly to the physiologically validated protein–protein interactome. TRS will also be suited to monitor the action of small-molecule inhibitors on protein–protein interactions in the living cell and hence could help to validate pharmacologically important targets. Furthermore, we have first promising indications that TRS can be the basis for the future development of a screening method for endogenous interacting proteins in mammalian cells in vivo, where endogenous proteins trans-SUMOylated by an expressed Ubc9-fusion protein could be purified and analysed by MS.

AUTHOR CONTRIBUTION

Ratnesh Srivastav designed and performed experiments and analysed data. Malte Klaus, Susan Schwede and Jessica Schwermann performed experiments. Rainer Niedenthal directed and supervised the research, designed the experiments, analysed data and wrote the paper. Matthias Gaestel co-supervised the research, analysed the data and wrote the paper.

FUNDING

This work was supported by the Institute for Physiological Chemistry/Biochemistry and the HiLF program of the Hannover Medical School (to R.N.).

Acknowledgments

We thank Astrid Jakobs for help with cell culture and Western blot experiments. We also thank Dr Alexey Kotlyarov (Hannover Medical School) for helpful discussions and Dr Thomas Binz (Hannover Medical School) for a critical reading of the paper before submission.

Abbreviations: BIFC, bimolecular complementation of a fluorescence signal; BRET, bioluminescence resonance energy transfer; CDC37, cell division cycle 37; CDK4, cyclin-dependent kinase 4; CRSP9, cofactor required for Sp1 transcriptional activation, subunit 9; CSNK2B, casein kinase 2, β polypeptide; DMEM, Dulbecco's modified Eagle's medium; ERK, extracellular-signal-regulated kinase; FKBP, FK506-binding protein; FRET, fluorescence resonance energy transfer; GFP, green fluorescent protein; EGFP, enhanced GFP; GST, glutathione transferase; HA, haemagglutinin; HDGF, hepatoma-derived growth factor; HEK, human embryonic kidney; HRP, horseradish peroxidase; HSF2BP, heat-shock factor 2-binding protein; Hsp, heat-shock protein; JNK, c-Jun N-terminal kinase; MAPK, mitogen-activated protein kinase; MAPPIT, Mammalian Protein–Protein Interaction Trap; MEK, MAPK/ERK kinase; MEKK1ca, constitutively active MEK kinase 1; MK, MAPK-activated protein kinase; PcG, polycomb group complex; PSMC3, proteasome 26S ATPase subunit 3-interacting protein; RFP, red fluorescent protein; ST, SUMOylation-site tag; STAT1, signal transducer and activator of transcription 1; SUMO, small ubiquitin-related modifier; TBP, TATA-box-binding protein; TAF10, TBP-associated factor 10; TEV, tobacco etch virus; TRS, trans-SUMOylation; UC, uncleavable; UFDS, Ubc9 fusion-directed SUMOylation; Y2H, yeast two-hybrid

References

View Abstract