REVIEW Methods for the detection and analysis of protein���protein interactions Tord Bergg��rd1, Sara Linse1 and Peter James2 1 Department of Biophysical Chemistry, Lund University, Lund, Sweden 2 Department of Protein Technology, Lund University, Lund, Sweden A large number of methods have been developed over the years to study protein���protein inter- actions. Many of these techniques are now available to the nonspecialist researcher thanks to new affordable instruments and/or resource centres. A typical protein���protein interaction study usually starts with an initial screen for novel binding partners. We start this review by describing three techniques that can be used for this purpose: (i) affinity-tagged proteins (ii) the two-hybrid system and (iii) some quantitative proteomic techniques that can be used in combination with, e.g., affinity chromatography and coimmunoprecipitation for screening of protein���protein inter- actions. We then describe some public protein���protein interaction databases that can be searched to identify previously reported interactions for a given bait protein. Four strategies for validation of protein���protein interactions are presented: confocal microscopy for intracellular colocaliza- tion of proteins, coimmunoprecipitation, surface plasmon resonance (SPR) and spectroscopic studies. Throughout the review we focus particularly on the advantages and limitations of each method. Received: February 6, 2007 Revised: April 12, 2007 Accepted: April 12, 2007 Keywords: Affinity tags / Protein complexes / Protein���protein interactions / Protein purification Proteomics 2007, 7, 2833���2842 2833 1 Introduction The human genome consists of 20 000���30 000 genes coding for over 500 000 different proteins of which more than 10 000 proteins can be produced by the cell at any given time (the cellular ���proteome���). Is has been estimated that over 80% of proteins do not operate alone but in complexes. These pro- tein���protein interactions are regulated by several mechan- isms. For example, metal-binding or PTMs can lead to con- formational changes that alter the affinity, co-operativity and kinetic parameters of the interaction. Many protein���protein interactions are part of larger cellular networks of protein��� protein interactions. It is believed that the cellular network of protein���protein interactions are built up by highly connected protein nodes (so called hubs) and many poorly connected nodes. Each node receives inputs and generates one or more specific outputs in a manner similar to computational units. Examples of important protein complexes are the spliceo- some, the ribosome and the nuclear pore complex. The basic architecture of the protein���protein interaction network is similar in all cells. Thus, hubs essential for cell survival are the same, but cell-specific differences can be found at the regulatory level. It has been shown previously that knock-out of a protein which has a central role in many networks tends to be lethal [1]. This phenomenon has been observed in many organisms [2���4] and is commonly referred to as the centrality-lethality rule. The current view that most, if not all, cellular proteins are directly or indirectly coupled in a large Correspondence: Dr. Tord Bergg��rd, Department of Biophysical Chemistry, Lund University, P. O. Box 124, SE-221 00 Lund, Swe- den E-mail: tord.berggard@bpc.lu.se Fax:146-46-2221495 Abbreviations: co-IP, coimmunoprecipitation SPR, surface plas- mon resonance TAP, tandem affinity purification TEV, tobacco etch virus Y2H, yeast two-hybrid DOI 10.1002/pmic.200700131 �� 2007 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.proteomics-journal.com

2834 T. Bergg��rd et al. Proteomics 2007, 7, 2833���2842 cellular protein���protein interaction network has implica- tions for the way we define cellular pathways. Thus, a given pathway (e.g., vesicle trafficking, apoptosis or cell cycle con- trol) can be regarded as a subsystem that is highly inter- connected to other pathways. We have during the last 10 years witnessed a tremendous development within the field of proteomics. Many innovative methods for the identification and characterization of pro- tein���protein interactions have been presented and many of them are currently in use in laboratories around the world. The technology development has in many cases been fuelled by the extraordinary advances in MS, which has made the identification of proteins a relatively simple task. In some cases macromolecular complexes such as ribosomes and exosomes have been purified and analysed directly in the mass spectrometer [5���8]. Thus mass spectrometric tech- niques can be used not only to identify individual proteins, but also to characterize biological assemblies. A number of large-scale studies have been presented, using e.g., two- hybrid screens [9���15] and coaffinity purification followed by MS [16���19] to detect protein���protein interactions on a ge- nome-wide scale. Somewhat surprisingly, only a small num- ber of the interactions are supported by more than one method [20]. Estimates of 40���80% false negatives and 30��� 60% false positives and have been assigned to high-through- put studies that have used two-hybrid techniques, affinity- based techniques or computational approaches [20���22]. The poor overlap can be explained partly by the fact that many different methods have been used. However, even within subsets of protein���protein interactions identified using the same method, the overlap can be poor (e.g., compare the results from the yeast two-hybrid (Y2H) screens from Ito et al. [10] and Uetz et al. [9]). It has been estimated that due to a high false-positive rate, current yeast and human interaction maps are roughly only 50 and 10% complete, respectively [23]. Although a number of methods have been developed lately to combat false-positive discoveries [24] it is clearly important to experimentally validate protein���protein inter- actions by several methods. In this review, we describe some methods that are commonly used to identify and verify pro- tein���protein interactions. We focus particularly on the advantages and disadvantages of each method. 2 Use of affinity tags for purification of protein complexes in vivo In vivo affinity fusion-based protein purification takes ad- vantage of the selective binding of a genetically fused affinity tag. First, cells are transfected with a plasmid coding for a bait protein fused to the tag. After an appropriate expression period, cells are lysed and the tagged bait, together with bound proteins, is isolated using a specific chemical or bio- logical ligand linked to a solid support. Eluted proteins are then separated by gel-electrophoresis and specifically bound proteins (i.e., proteins absent from the control) are identified by MS. Compared with other techniques, fusion-based affin- ity protein purification is an excellent method to purify and identify multiprotein complexes. Many other widely used techniques, such as the Y2H system, cannot detect interac- tions involving more than two proteins (see ref. [25]). As the tagged protein is expressed in vivo it can undergo PTMs. This has important implications for the purification of protein complexes from mammalian cells because PTMs, such as phosphorylation, are often used by regulatory pro- teins to increase or decrease the affinity for their target pro- teins. Furthermore, two proteins that interact in vitro may not be expressed in the same cells (or cellular subcompart- ments) and therefore the interaction observed in vitro may be nonphysiological. By expressing the bait protein directly in cells it is allowed to be directed to its correct subcellular location and to associate with its physiological targets. How- ever, during cell lysis, the tagged protein is present in a mix- ture of both physiological and nonphysiological targets. Therefore, nonphysiological targets may sometimes be pres- ent in the lysate and may be incorporated in the complex. This may be a problem, especially if the physiological com- plex is composed of proteins interacting with fast kinetics (high Kon, high Koff ) and/or low affinity, and thus more easily replaced by nonphysiological ones. Affinity-based methods are biased towards proteins that interact with high affinity and with slow kinetics of dissocia- tion. Thus affinity chromatography-based procedures may not be optimal for the detection of transient protein interac- tions, especially if stringent rinsing procedures are used. For example, transient complexes involved in post-translational control of protein activity, may escape detection. What is the reason for the bias towards high affinity interacting proteins? First, one must consider that the intracellular milieu is very different from that in the test tube. Protein���protein interac- tions are not designed to occur in dilute buffers such as those that researchers use in the laboratory. Instead all protein��� protein interactions inside a cell occur in a concentrated mixture of macromolecules. In fact the protein concentra- tion in the cytosol may be as high as that of some protein crystals [26]. The high intracellular protein concentration influences the rate at which molecules diffuse in the cell and it leads to competition for water (referred to as macro- molecular crowding). As a consequence, the binding affinity between proteins in a complex may be much higher in the crowded environment inside a cell compared to two proteins in a buffer. This may affect the retrieval of weakly interacting proteins in experiments relying on affinity-based purification of tagged proteins: It is possible that weakly interacting pro- teins are indeed bound to the tagged protein inside the cell, but dissociate after cell lysis and affinity purification of the tagged protein. Surprisingly, the fact that affinities between interacting proteins may be higher in vivo than in vitro (because of macromolecular crowding effects) has not been discussed very much in the literature on methods for purification and identification of protein���protein inter- actions. �� 2007 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.proteomics-journal.com