Two s company, three s a crowd: the yeast two hybrid system for mapping molecular interactions Emma Warbrick

Ways & Means 13 Two s company, three s a crowd: the yeast two hybrid system for mapping molecular interactions Emma Warbrick Address: Department of Anatomy and Physiology, Medical Sciences Institute, University of Dundee, Dundee DD1 4HN, UK. E-mail: e.warbrick@dundee.ac.uk Electronic identifier: 0969-2126-005-00013 Structure 15 January 1997, 5:13 17 Current Biology Ltd ISSN 0969-2126 Introduction Use of the yeast two hybrid system as a screening method for the detection of novel protein protein interactions is an increasingly popular and useful technique. Many reviews have already been written concerning the background and development of the systems used [1 6]. Here I hope to answer some of the most commonly asked questions concerning the applications of the two hybrid system, look at its advantages and disadvantages, and describe some of the newer applications of the system including the so-called three hybrid and reverse two hybrid approaches. Nuts and bolts The two hybrid system takes advantage of the modular domain structure of eukaryotic transcription factors: transcriptional activation and sequence-specific DNA-binding regions exist as separable domains which need to be physically associated, though not necessarily covalently linked, for reconstitution of activity [7]. In the two hybrid system, activation of reporter constructs occurs when the two domains are brought together through the interaction of two polypeptides expressed as activation domain and DNAbinding domain fusion proteins (Fig. 1) [8,9]. Two hybrid systems all require the expression of these two fusion proteins from specially designed vectors: one encodes the sequence-specific DNA-binding domain, such as that from the yeast transcription factor GAL4, or the bacterial repressor protein LexA. The second vector encodes the transcription activation domain, usually from either GAL4 or the herpes simplex virus protein VP16. Although strong activation domains should, in principle, allow detection of weaker interactions, their expression can also be deleterious to the cell due to generalized effects on transcription, such as squelching. These fusion proteins are expressed either from a strong constitutive promoter, such as ADH1, or from a conditional promoter; the latter may be advantageous if for any reason constructs are deleterious to yeast growth. In library screening experiments, the protein used for screening is usually expressed as a DNA-binding hybrid (the bait ), while the plasmids of the cdna library express activation domain fusion proteins (the prey ). The third essential component is the yeast strain in which the interaction assays will be performed. This must fulfil certain requirements, including auxotrophic mutations for the maintenance by growth selection of the various plasmids involved and, in GAL4-based two hybrid systems, deletion of genes encoding wild type GAL4 and the GAL4-binding protein GAL80. It is worth noting that GAL4 deletion strains cannot be used in conjunction with conditional GAL1 promoters which require wild type GAL4 protein. The yeast strain also contains appropriately regulated reporter constructs controlled either by the LexA operator, or a GAL4-binding domain (e.g. from GAL1), which are either integrated into the genome or present as multicopy plasmids. The reporter construction determines its sensitivity; this in turn will determine the stringency with which interactions will be detected and so is an important factor in setting up screens. Reporters can differ in the number and affinity of upstream binding sites for the bait, and in the position of these sites relative to the start of transcription. All these factors can affect the strength of protein interaction detected. The most commonly used reporter is the lacz gene, which encodes the enzyme -galactosidase. The expression of lacz can be monitored using a colorimetric assay based on the activity of -galactosidase, and is useful for visual assays of binding activity on plates containing 3-bromo-4-chloro-3-indoyl- -galactoside (X-gal), or in in vitro assays as an indication of the reporter activity. Other commonly used reporters are yeast genes, such as HIS3 or LEU2, which can give a growth selection for interaction and so are very useful for screening large numbers of yeast colonies for interacting clones. A variant of the yeast two hybrid system has been used in mammalian cells. Various reporters have been used, including the gene encoding chloramphenicol resistance (CAT), the cell surface marker CD4, and the resistance gene for Hygromycin B, which allows growth selection [10]. This technique may prove to be most useful for the study of protein protein interactions which are dependent upon protein modifications, or specific conformational states which do not occur in yeast. Commonly asked questions How easy is it to do technically? Very few specialized techniques are required for two hybrid work, though a basic knowledge of yeast culture,

14 Structure 1997, Vol 5 No 1 Figure 1 (a) Two hybrid interaction (b) Disruption (c) Competition (d) Three hybrid interaction DNA-binding domain hybrid and Bridging protein TRANSCRIPTION Reportergene NO TRANSCRIPTION NO TRANSCRIPTION TRANSCRIPTION Competitor protein Transcriptional activation domain hybrids Four examples of two and three hybrid approaches. (a) The DNAbinding domain fusion protein (the bait) specifically binds the upstream activation sequence () in the reporter construct. The interaction of the activation domain fusion protein (the prey) with the bait results in the expression of the reporter construct. (b) The plasmid expressing the activation domain fusion protein has been subject to mutagenesis, and here we see a protein mutated in the protein-interaction domain. As the two fusion proteins can no longer interact, reporter gene expression is not turned on. (c) Expression of a third protein which binds strongly to the bait results in the dissociation of the bait prey complex; as above, reporter expression is turned off. Growth of yeast colonies where interactions are disrupted can be directly selected for by the use of a counter-selectable marker. (d) In this three-way interaction, the interaction between bait and prey is mediated by, and dependent upon, expression of a third, bridging protein. and an appreciation of yeast genetic techniques is a considerable advantage; I would strongly recommend consulting your local Saccharomyces cerevisiae expert. Will my protein be suitable for analysis in this system? The most common reason for a protein being unsuitable for two hybrid analysis is that it activates transcription when expressed alone, either as a DNA-binding or activation domain hybrid. It is very difficult to predict whether this will be the case, and unfortunately the only definitive answer will come from making a construct, and testing it in yeast. If the protein does, for instance, activate transcription strongly, the only answer may be to test truncated versions. For mapping interaction domains this may not be a problem, but screening libraries with a truncated protein that may not have its full binding capabilities may not be such an attractive prospect. How long will it take and is it time consuming? The length of time required for two hybrid screening can be very variable: unfortunately false positives nearly always occur and must be excluded. A two hybrid screen will take a couple of months from beginning to end, though this is obviously dependent upon how many positive colonies you pick to analyze! The simple testing of one protein against another will take considerably less time. Two hybrid experiments begin intensively with the construction of expression constructs, but yeast transformation and the analysis of results is less intensive. If you are screening libraries, the isolation of plasmids, sequencing and retransformation to check for false positives can be a full time occupation. Where can I get the components from and will it be expensive? Two hybrid systems, including libraries, are now available commercially, though they may also be available through academic contacts. Yeast culture and transformation are not very expensive procedures; the cost will be approximately the same as that of work with Escherichia coli. The only specialist pieces of equipment required are a temperature controlled incubator and shaker for yeast cultures. Possibly the most expense will be incurred if you buy vectors, and especially libraries, commercially. Are there any tricks I can use to save myself a lot of work? Plasmids containing a range of antibiotic markers have been described which ease the identification of plasmids rescued into E. coli [11]. An alternative is to identify plasmids by complementation of auxotrophic mutations in E. coli: for instance, LEU2 complements leub6, and TRP1 complements trpc. Recently, it has become feasible to use mating assays in which plasmids are tested against each other by mating yeast strains containing the relevant plasmids, rather than using co-transformation [4,12]. These methods not only save time in routine work, but make extended analysis of protein networks possible (see below).

Ways & Means Yeast two hybrid mapping Warbrick 15 Will I get a good result out at the end? Screening in the two hybrid system is unpredictable, so you must accept the risk of a negative result. Use of the two hybrid system for the analysis of single protein protein interactions gives more straightforward results, although a positive result will have to be backed up by biochemical data, preferably in vivo, for it to be acceptable for publication. Negative results are more difficult to interpret: false negatives are possible for a variety of reasons (e.g. protein instability, problems with nuclear localization and toxic effects to the yeast cell). Some of these possibilities are testable, for example protein expression can be examined by Western blotting. The availability of antibodies need not be a problem: many two hybrid vectors express epitope tags, or GAL4/LexA monoclonal antibodies are available. To test whether a LexA fusion protein reaches the nucleus a repressor assay can be used: a reporter construct is used which expresses LacZ from the GAL1 promoter, but which contains a LexA operator between the GAL1 upstream activation sequence and lacz [4,13]. Binding of the LexA fusion protein results in a decrease in LacZ expression. Another problem that is not widely discussed is steric hindrance: if two proteins interact across a wide interface, it is possible that the activation domain may be sterically prevented from activating transcription. Applications Screening libraries This form of analysis has been widely used to great effect to detect novel protein protein interactions. It has the advantages that protein protein interactions may be detected in a eukaryotic cellular environment and many potential interactors can be rapidly and simultaneously screened. The only requirements are a cdna clone of the bait protein, and a suitable cdna library for screening. No additional reagents, such as high specificity antibodies or specialized protein purification techniques are required. Once positive results are obtained, DNA sequencing of the library plasmids will provide information concerning the primary structure, and possibly the identity, of the interacting protein. The isolation of truncated fusion proteins can give immediate information about interaction domains. As the level of reporter activation correlates well with the specific binding activities, measurement of reporter activity gives an immediate indication of the strength of the interaction [14]. The binding of bait proteins to wild type and mutant versions of your prey can be compared: for example, Iwabuchi et al. have used this method to identify proteins that bind to wild type, but not mutant forms of p53 [15]. The most common problem associated with this technique is the identification false positives. These are fusion proteins that activate transcription when expressed alone, or show a high level of non-specific interactions. A very useful table of commonly identified two hybrid false positives has been compiled (http://www.fccc.edu:80/research/labs/ golemis/interactiontrapinwork; a web page which gives details of false positives identified in two hybrid screens). The elimination of false positives requires extensive control experiments: clones must be tested alone and in combination with a number of unrelated bait fusion proteins. A popular criticism aimed against two hybrid screening experiments is that it detects protein interactions that do not necessarily occur in vivo. It is worth remembering that nearly all methods for analyzing protein interactions can result in artefacts! Analysis of protein interaction domains Once an interaction between two proteins in the two hybrid system has been established, then specific interaction domains can be investigated. This form of analysis, as described above, is not technically difficult and does not require specialized reagents, though it may not yield such rapid results as in vitro analysis. In a simple approach, subdomains of proteins may be expressed, or deleted, and the resulting constructs tested for interactions with known partners; in this way it can be possible to identify distinct interaction domains. For example, this approach has been successful in defining the region of proliferating cell nuclear antigen (PCNA) essential for its interaction with the CDK inhibitor p21 Cip1, results which have recently been confirmed by detailed structural studies [16,17]. Studies of homodimerization can very conveniently be carried out in the two hybrid system as specialist labelling or tagging of proteins is not required. In one of the founding two hybrid system papers, Chien et al. extensively studied homodimerization of the SIR4 protein, and identified the small region required for this process [9]. Another advantage of the system lies in the genetic aspect of the two hybrid approach: mutant plasmids can be screened for changes that disrupt the protein protein interaction by looking for loss of reporter expression (see reverse two hybrid below). Such an approach has been used to screen for p53 mutants which can no longer bind to SV40 large T antigen [18]. Reymond and Brent have used the two hybrid system in a study of the cyclin-dependent kinase (CDK) binding capacity of p16 allelic variants detected in melanoma-prone families [19]. Peptide protein interactions Combinatorial libraries are being increasingly used as a means for identifying ligands in biologically important interactions, and peptide libraries in particular have the advantage that they can be expressed in vivo. Phage display, in which a peptide motif is expressed on the surface of filamentous phage is widely used [5]. Once such a phage library has been screened by sequential rounds of affinity enrichment, the primary structure of the displayed ligand can be identified by DNA sequencing of the phage

16 Structure 1997, Vol 5 No 1 DNA. In these forms of analysis, the DNA is acting as a molecular tag for the primary sequence of the ligand. Methods for screening peptide libraries for interactions in the two hybrid system are now available, these have unique features which make them a valuable form of screening. This form of two hybrid screening has the advantage that peptides may be expressed in a variety of contexts. Peptides have been expressed directly fused to the GAL4 activation domain, or as structurally constrained motifs displayed as part of a larger protein [20,21]. The E. coli protein thioredoxin (TrxA) has been used in this way: it is small, soluble and readily expressed in E. coli, thus providing a readily usable biochemical tool. This approach has all the advantages associated with two hybrid analysis, and particularly has the advantage over phage display screening in that a single round of screening allows the direct detection of clones with varying binding affinities, as reflected in levels of reporter gene expression. Once interacting peptides have been identified, then they are amenable to genetic approaches, such as mutagenic PCR, to identify residues essential for the interaction. In one study, seven peptides capable of binding the product of the Retinoblastoma tumour suppressor gene (Rb) were identified: all contained a common motif, which was further analyzed by mutagenesis [20]. Colas et al. selected CDK2-binding peptides expressed within TrxA [21]. These peptides showed distinct patterns of cross-reactivity with different CDKs, although none shared specific motifs. These peptide aptimers seem to share some of the characteristics of monoclonal antibodies. The expression of peptides may also be used in the analysis of the dissociation of protein protein interactions (see below). Reverse two hybrid analysis Using a reverse two hybrid approach, positive selection can be imposed for dissociating mutations in interacting proteins (Fig. 1) [6,22,23]. One approach involves the use of the yeast URA3 gene as a reporter. Not only is the URA3 gene product essential for uracil biosynthesis, it also catalyzes the conversion of the artificial substrate 5-fluoroorotic acid (5-FOA) into a toxic compound. Thus plasmids expressing interacting two hybrid fusion proteins can be mutagenized (separately or simultaneously), introduced into yeast and colonies where the interaction no longer takes place selected for by the inclusion of 5-FOA in the medium. Vidal et al. have used this system to identify mutations in the transcription factor E2F1 that disrupt heterodimerization with its binding partner DP1 [23]. They showed it was possible to exclude relatively uninteresting mutations, such as truncations and frame shifts, by employing a secondary screen using a very sensitive HIS3 reporter gene, which identified clones retaining a very weak DP1- binding capacity. For strongly dissociating mutations, it was necessary to screen the E2F1 mutants for their ability to bind Rb, which interacts with the C terminus of E2F. The application of this form of reverse technology to library screening opens the door to large scale experiments which previously were unfeasible. Detecting three-way interactions This variant of two hybrid analysis can be used to examine the formation of heterotrimeric complexes. Reporter expression is dependent on the assembly of activation and DNAbinding domains, but in this case the interaction is mediated by a bridging protein (Fig. 1) [24]. In this example, the DNA-binding domain and transcriptional activation domain fusion proteins interact independently with separate domains of a third protein, a bridging protein, though not with each other; reporter activation is, therefore, dependent upon co-expression of the bridging protein. This very elegant way to demonstrate a three-way interaction is limited to fusion proteins that do not interact and to bridging proteins that do not multimerize, otherwise the two fusion proteins could interact with separate bridging molecules. Van Aelst et al. first used this technique to demonstrate that Ras and the protein kinase MEK have distinct binding domains within the Raf oncogene, which is also a kinase, and that the three can form a complex [24]. A variant of this system using an RNA bridging molecule has also been used [25]. Screening for the dissociation of protein interactions Recently the ideas of counter selection and the three hybrid approach have been combined in the development of screens for molecules that induce specific protein protein dissociations (Fig. 1) [6,22,23]. The URA3 reporter gene is used counterselectively, so where expression of a third protein results in the dissociation of the two hybrid pair, 5-FOA resistance is conferred on the transformed cells. Vidal et al. showed the feasibility of such experiments using the adenovirus protein E1A, which is known to disrupt the interaction of E2F with Rb and p107 in vivo, and proposed that either peptide or protein libraries could be screened for dissociating molecules [22,23]. It should be possible to identify not only specific competitors of protein interactions, but also molecules which induce structural changes in either interacting partner that result in dissociation. In all diseases attributed to a particular protein protein interaction, specific dissociation can be viewed as a potential therapeutic strategy. The identification of peptides capable of dissociating these interactions will be valuable as lead compounds for drug discovery programs. Analysis of protein interaction networks Using mating assays, pairs of two hybrid plasmids can be brought together by crossing yeast strains and analyzing the diploid progeny, rather than by using co-transformation. This method was originally devised to facilitate the testing of prey plasmids against a variety of controls [12]. Interaction mating has now been used to examine interactions

Ways & Means Yeast two hybrid mapping Warbrick 17 between small sets of proteins. Finley and Brent used this approach to test a set of cyclin-dependent kinase (CDK) interactors for interaction with a range of CDKs [26]. This approach has been extended to library screening: the library is stored as a bank of yeast transformants which is mated to a strain containing the bait plasmid, the resulting progeny are screened for reporter expression [27]. The mating assay has proved to be an essential tool in the use of the two hybrid system to analyze protein interaction networks [28]. In the first example of this approach, Bartel et al. have applied two hybrid technology to characterize protein interactions within an entire genome [29]. In order to identify binary interactions between the bacteriophage T7 proteins they screened two genomic libraries against each other: one expressing activation domain hybrids the other expressing DNA-binding domain hybrids. This was done initially in pools, followed by individual testing. An unexpected finding of these experiments was the identification of intramolecular associations. This form of interaction has not previously been widely detected, presumably because predominantly full length proteins, in which intramolecular domains are not revealed, are used for interaction screening. The two hybrid assay could thus be used to complement biochemical and biophysical analyses of protein folding and structure. The scaling up of such mapping experiments to give protein interaction maps for an entire organism is an exciting prospect. Conclusions From the seminal observation made by Fields and Song in 1989, the two hybrid system has rapidly developed into an incredibly robust and versatile method for the study of protein protein interactions. If progress in the next eight years matches that which has been made up until now, then the prospects for the application of this form of analysis are mind-boggling. Acknowledgements I would like to thank all of my colleagues for their helpful suggestions, and to acknowledge the support of the Cancer Research Campaign. References 1. Bartel, P.L., Chien, C.-T., Sternglanz, R. & Fields, S. (1993). Using the two-hybrid system to detect protein protein interactions. In Cellular Interactions in Development: a Practical Approach. (Hartley, D.A., ed.), pp. 153 179, Oxford University Press, Oxford, UK. 2. Fields, S. & Sternglanz, R. (1994). The two-hybrid system: an assay for protein protein interactions. Trends Genet. 10, 286 292. 3. Mendelsohn, A.R. & Brent, R. (1994). Applications of interaction traps/two hybrid systems to biotechnology research. Curr. Opin. Biotech. 5, 482 486. 4. Finley, R.L. & Brent, R. (1995). Interaction trap cloning with yeast. In DNA Cloning 2, Expression Systems: a Practical Approach. (Hames, B.D. & Glover, D.M., eds), pp. 169 203, Oxford University Press, Oxford, UK. 5. Allen, J.B., Walberg, M.W., Edwards, M.C. & Elledge, S.J. (1995). Finding prospective partners in the library: the two hybrid system and phage display find a match. Trends Biochem. Sci. 20, 511 516. 6. White, M.A. (1996). The yeast two hybrid system: forward and reverse. Proc. Natl. Acad. Sci. USA 93, 10001 10003. 7. Ma, J. & Ptashne, M. (1987). A new class of transcriptional activators. Cell 51, 113 119. 8. Fields, S. & Song, O. (1989). A novel genetic system to detect protein protein interactions. Nature 340, 245 247. 9. Chien, C.T., Bartel, P.L., Sternglanz, R. & Fields, S. (1991). The twohybrid system: a method to identify and clone genes for proteins that interact with a protein of interest. Proc. Nat. Acad. Sci. USA 88, 9578 9582. 10. Fearon, E.R., et al., & Dang, C.V. (1992). Karyoplasmic interaction selection strategy: a general strategy to detect protein protein interaction in mammalian cells. Proc. Nat. Acad. Sci. USA 89, 7958 7962. 11. Watson, M.A., Buckholz, R. & Weiner, M.P. (1996). Vectors encoding alternative antibiotic resistance for use in the yeast two-hybrid system. Biotechniques 21, 255 259. 12. Harper, J.W., Adami, G.R., Wei, N., Keyomarsi, K. & Elledge, S.J. (1993). The p21 Cdk-interacting protein Cipl is a potent inhibitor of Gl cyclin-dependent kinases. Cell 75, 805 816. 13. Brent, R. & Ptashne, M. (1984). A bacterial repressor protein or yeast transcriptional terminator can block upstream activation of a yeast gene. Nature 312, 612 615. 14. Estojak, J., Brent, R. & Golemis, E.A. (1995). Correlation of two-hybrid affinity data with in vitro measurements. Mol. Cell. Biol. 15, 5820 5829. 15. Iwabuchi, K., Bartel, P.L., Li, B., Bartel P.L., Marraccino, R. & Fields, S. (1994). Two cellular proteins that bind to wild type but not mutant p53. Proc. Natl. Acad. Sci. USA 91, 6098 6102. 16. Warbrick, E., Lane, D.P., Glover, D.M. & Cox, L.S. (1995). A small peptide inhibitor of DNA replication defines the site of interaction between the cyclin dependent kinase inhibitor p21 WAF1 and proliferating cell nuclear antigen. Curr. Biol. 5, 275 282. 17. Gulbis, J.M., Kelman, Z., Hurwitz, J., O Donnell, M. & Kuriyan, J. (1996). Structure of the C-terminal region of p21 WAF1/CIP1 complexed with human PCNA. Cell 87, 1 20. 18. Li, B. & Fields, S. (1993). Identification of mutations in p53 that affect its binding to SV40 large T antigen by using the yeast two-hybrid system. FASEB J. 7, 957 963. 19. Reymond, A. & Brent, R. (1995). p16 proteins from melanoma-prone families are deficient in binding to cdk4. Oncogene 11, 1173 1178. 20. Yang, M., Wu, Z. & Fields, S. (1995). Protein peptide interactions analyzed with the yeast two-hybrid system. Nucleic Acids Res. 23, 1152 1156. 21. Colas, P., Cohen, B., Jessen, T., Grishina, I., McCoy, J. & Brent, R. (1996). Genetic selection of peptide aptimers that recognize and inhibit cyclin-dependent kinase 2. Nature 380, 548 550. 22. Vidal, M., Brachmann, R.K., Fattaey, A., Harlow, E. & Boeke, J.D. (1996). Reverse two-hybrid and one-hybrid systems to detect dissociation of protein protein and DNA protein interactions. Proc. Natl. Acad. Sci. USA 93, 10315 10320. 23. Vidal, M., Braun, P., Chen, E., Boeke, J.D. & Harlow, E. (1996). Genetic characterisation of a mammalian protein protein interaction domain by using a reverse two-hybrid system. Proc. Natl. Acad. Sci. USA 93, 10321 10326. 24. Van Aelst, L., Barr, M., Marcus, S., Polverino, A. & Wigler, M. (1993). Complex formation between RAS and RAF and other protein kinases. Proc. Natl. Acad. Sci. USA 90, 6213 6217. 25. SenGupta, D.J., Zhang, B., Kraemer, B., Pochart, P., Fields, S. & Wickens, M. (1996). A three hybrid system to detect RNA protein interactions in vivo. Proc. Natl. Acad. Sci. USA 93, 8496 8501. 26. Finley, R.L. & Brent, R. (1994). Interaction mating reveals binary and ternary connections between Drosophila cell cycle regulators. Proc. Natl. Acad. Sci. USA 91, 12980 12984. 27. Bendixen, C., Gangloff, S. & Rothstein, R. (1994). A yeast matingselection scheme for detection of protein protein interactions. Nucleic Acids Res. 22, 1778 1779. 28. Finley, R.L. & Brent, R. Two-hybrid analysis of genetic regulatory networks. In The Yeast Two Hybrid System. (Bartel, P.L. & Fields, S., eds), Oxford University Press, Oxford, in press. 29. Bartel, P.L., Roecklein, J.A., SenGupta, D. & Fields, S. (1996). A protein linkage map of Escherichia coli bacteriophage T7. Nat. Genet. 12, 72 77.