MICROBIOLOGICAL REVIEWS, Mar. 1995, p. 94���123 Vol. 59, No. 1 0146-0749/95/$04.0010 Copyright q 1995, American Society for Microbiology Protein-Protein Interactions: Methods for Detection and Analysis ERIC M. PHIZICKY1* AND STANLEY FIELDS2 Department of Biochemistry, University of Rochester Medical School, Rochester, New York 14642,1 and Department of Molecular Genetics and Microbiology, State University of New York at Stony Brook, Stony Brook, New York 117942 INTRODUCTION.........................................................................................................................................................95 PHYSICAL METHODS TO SELECT AND DETECT PROTEINS THAT BIND ANOTHER PROTEIN ......96 Protein Affinity Chromatography ...........................................................................................................................96 Purity of the coupled protein and use of protein fusions...............................................................................97 Influence of modification state............................................................................................................................97 Retention of native structure of the coupled protein ......................................................................................97 Concentration of the coupled protein................................................................................................................98 Amount of extract applied...................................................................................................................................98 Other considerations............................................................................................................................................98 Affinity Blotting.........................................................................................................................................................98 Immunoprecipitation................................................................................................................................................99 Cross-Linking..........................................................................................................................................................100 Determination of architecture...........................................................................................................................100 Detection of interacting proteins......................................................................................................................101 (i) Detection in vivo........................................................................................................................................101 (ii) Detection in vitro .....................................................................................................................................101 (iii) Other considerations..............................................................................................................................102 LIBRARY-BASED METHODS.................................................................................................................................102 Protein Probing.......................................................................................................................................................102 Phage Display..........................................................................................................................................................103 Basic approach....................................................................................................................................................103 Related methods..................................................................................................................................................104 (i) Antibody phage..........................................................................................................................................104 (ii) Peptides on plasmids...............................................................................................................................105 Two-Hybrid System.................................................................................................................................................105 Other Library-Based Methods..............................................................................................................................106 GENETIC METHODS...............................................................................................................................................106 Extragenic Suppressors..........................................................................................................................................107 Synthetic Lethal Effects .........................................................................................................................................107 Overproduction Phenotypes ..................................................................................................................................108 Overproduction of wild-type proteins ..............................................................................................................108 Overproduction of mutant proteins .................................................................................................................108 Unlinked Noncomplementation.............................................................................................................................109 POPULAR METHODS TO ESTIMATE AND DETERMINE BINDING CONSTANTS .................................109 Importance of Characterization of the Binding Interaction.............................................................................109 Binding constant.................................................................................................................................................109 Concentrations of species ..................................................................................................................................109 Influence of competing proteins........................................................................................................................109 Influence of cofactors .........................................................................................................................................109 Effect of cellular compartmentation.................................................................................................................109 Solution conditions.............................................................................................................................................109 Limits of Binding-Constant Considerations .......................................................................................................109 Methods for Determining Binding Constants ....................................................................................................110 Binding to immobilized proteins ......................................................................................................................110 Sedimentation through gradients.....................................................................................................................111 Gel filtration columns ........................................................................................................................................111 (i) Nonequilibrium ������small-zone������ gel filtration columns...........................................................................111 (ii) Hummel-Dreyer method of equilibrium gel filtration.........................................................................111 (iii) Large-zone equilibrium gel filtration...................................................................................................112 Sedimentation equilibrium................................................................................................................................112 * Corresponding author. Mailing address: Department of Biochem- istry, University of Rochester Medical School, 601 Elmwood Ave., Box 607, Rochester, NY 14642. Phone: (716) 275-7268. Fax: (716) 271-2683. 94

Fluorescence methods ........................................................................................................................................112 (i) Fluorescence spectrum .............................................................................................................................113 (ii) Fluorescence polarization or anisotropy with tagged molecules.......................................................113 Solution equilibrium measured with immobilized binding protein.............................................................114 Surface plasmon resonance...............................................................................................................................114 Limits to Detection.................................................................................................................................................115 EXAMPLES OF WELL-CHARACTERIZED DOMAINS.....................................................................................115 Leucine Zipper ........................................................................................................................................................116 Structure ..............................................................................................................................................................116 Stability ................................................................................................................................................................116 Specificity.............................................................................................................................................................116 Regulation............................................................................................................................................................116 SH2 Domain ............................................................................................................................................................116 SH3 Domain ............................................................................................................................................................117 CONCLUDING REMARKS......................................................................................................................................118 ACKNOWLEDGMENTS ...........................................................................................................................................118 REFERENCES ............................................................................................................................................................118 INTRODUCTION Protein-protein interactions are intrinsic to virtually every cellular process. Any listing of major research topics in biolo- gy���for example, DNA replication, transcription, translation, splicing, secretion, cell cycle control, signal transduction, and intermediary metabolism���is also a listing of processes in which protein complexes have been implicated as essential components. In consequence, the analysis of the proteins in these complexes is no longer the exclusive domain of biochem- ists geneticists, cell biologists, developmental biologists, mo- lecular biologists, and biophysicists have by necessity all gotten into the act. We attempt in this review to summarize both classical and recent methods to identify proteins that interact and to assess the strengths of these interactions. Proteins that are composed of more than one subunit are found in many different classes of proteins. Some of the best- characterized multisubunit proteins are those that, as originally purified, contained two or more different components. These include classical proteins such as hemoglobin, tryptophan syn- thetase, aspartate transcarbamylase, core RNA polymerase, Qb-replicase, and glycyl-tRNA synthetase. Since these pro- teins purified as multisubunit complexes, their protein-protein interactions were self-evident. Other well-known examples of multisubunit proteins include much more complicated assemblies of polypeptides. These in- clude metabolic enzymes such as the pyruvate dehydrogenase and a-ketoglutarate dehydrogenase complexes, the DNA rep- lication complex of Escherichia coli and other organisms, the bacterial flagellar apparatus, the nuclear pore complex, and the tail assembly of bacteriophage T4. Also included in this group are ribonucleoprotein complexes, such as the signal recogni- tion particle of the glycosylation pathway, small nuclear ribo- nucleoproteins of the spliceosome, and the ribosome itself. Although some of the subunits of these protein complexes are not tightly bound, activity is associated with a large structure that in many cases is called a protein machine (5). There are also a large number of transient protein-protein interactions, which in turn control a large number of cellular processes. All modifications of proteins necessarily involve such transient protein-protein interactions. These include the interactions of protein kinases, protein phosphatases, glycosyl transferases, acyl transferases, proteases, etc., with their sub- strate proteins. Such protein-modifying enzymes encompass a large number of protein-protein interactions in the cell and regulate all manner of fundamental processes such as cell growth, cell cycle, metabolic pathways, and signal transduction. Transient protein-protein interactions are also involved in the recruitment and assembly of the transcription complex to spe- cific promoters, the transport of proteins across membranes, the folding of native proteins catalyzed by chaperonins, indi- vidual steps of the translation cycle, and the breakdown and re-formation of subcellular structures during the cell cycle (such as the cytoplasmic microtubules, the spindle apparatus, nuclear lamina, and the nuclear pore complex). Transient com- plexes are much more difficult to study, because the proteins or conditions responsible for the transient reaction have to be identified first. Part of the goal of this review is to describe recent methods and developments that have allowed their identification and characterization. Protein-protein interactions can have a number of different measurable effects. First, they can alter the kinetic properties of proteins. This can be reflected in altered binding of sub- strates, altered catalysis, or (as first enunciated by Monod et al. [153]) altered allosteric properties of the complex. Thus, the interaction of proliferating-cell nuclear antigen with DNA polymerase d alters the processivity of the polymerase (174), the interaction of succinate thiokinase and a-ketoglutarate de- hydrogenase lowers the Km for succinyl coenzyme A by 30-fold (171), and the cooperative binding of oxygen to hemoglobin and the allosteric regulation of aspartate transcarbamylase are regulated by interactions of the protomers. Second, protein- protein interactions are one common mechanism to allow for substrate channeling. The paradigm for this type of complex is tryptophan synthetase from Neurospora crassa. It is a complex of two subunits, each of which carries out one of the two steps of reaction (formation of indole from indole 3-glycerol phos- phate, followed by conversion of indole to tryptophan). The intermediate indole is noncovalently bound, but it is preferen- tially channeled to form tryptophan (241). Many similar exam- ples of metabolic channeling have been demonstrated, both between different subunits of a complex and between different domains of a single multifunctional polypeptide (see reference 208 for a review). Third, protein-protein interactions can result in the formation of a new binding site. Thus, an ADP site forms at the interface of the a and b subunits of Escherichia coli F1-ATPase (228), yeast hexokinase binds one ATP molecule at the interface of the asymmetric homodimer (209), and phos- phofructokinase from Bacillus stearothermophilus binds both fructose 6-phosphate and ADP at the interface between sub- units (60). Fourth, protein-protein interactions can inactivate a protein this is the case with the interaction of phage P22 repressor with its antirepressor (213), with the interaction of trypsin with trypsin inhibitor (221), and with the interaction of VOL. 59, 1995 PROTEIN-PROTEIN INTERACTIONS 95

phage T7 gene 1.2 protein with E. coli dGTP triphosphohy- drolase (156). Fifth, protein-protein interactions can change the specificity of a protein for its substrate thus, the interaction of lactalbumin with lactose synthase lowers the Km for glucose by 1,000-fold (95), and the interaction of transcription factors with RNA polymerase directs the polymerase to different pro- moters. Klotz et al. (116) enumerated four advantages of multisub- unit proteins relative to a single large protein with multiple sites. First, it is much more economical to build proteins from simpler subunits than to require multiple copies of the coding information to synthesize oligomers. Thus, for example, actin filaments and virus coats are much more simply assembled from monomers than by translation of a large polyprotein of repeated domains. Similarly, it is much more convenient to have one gene encoding a protein with different interacting partners, such as some of the eukaryotic RNA polymerase subunits, than to have the gene for that subunit reiterated for each different polymerase. Second, translation of large pro- teins can cause a significant increase in errors in translation if such errors cause a lack of activity, they are much more eco- nomically eliminated by preventing assembly of that subunit into the complex than by eliminating the whole protein. Third, multisubunit assemblies allow for synthesis at one locale, fol- lowed by diffusion and assembly at another locale this allows for both faster diffusion (since the monomers are smaller) and compartmentalization of activity (if assembly is required for activity). Fourth, homooligomeric proteins, if they have an advantage over monomers, are easily selected in evolution if the oligomers interact in an antiparallel arrangement in this case, a single-amino-acid change that increases interaction po- tential has effects at two such sites. Another advantage of multisubunit complexes is the ability to use different combinations of subunits to alter the magni- tude or type of response. Thus, for example, adult hemoglobin (a2b2) and fetal hemoglobin (a2g2) are each composed of heterooligomers with a common a subunit differences in the binding of oxygen in these hemoglobins allow oxygen to be readily passed from mother to fetus. Other examples include the oligomerization of Jun with Fos or with itself, which results in distinct activities in transcription because the different dimers bend DNA in opposite directions (114) the interaction of TATA-binding protein with the transcription apparatus of RNA polymerase I, II, or III, in which TATA-binding protein plays different roles (235) the interactions of microtubules with the large set of proteins to which they bind (113), not all of which bind at the same time the interaction of different transcription factors with core RNA polymerase in both eu- karyotes and prokaryotes to direct transcription of different genes and the interaction of retinoblastoma (Rb) protein with viral oncoproteins and other cellular proteins (31, 32). Protein-protein interactions may be mediated at one ex- treme by a small region of one protein fitting into a cleft in another protein and at another extreme by two surfaces inter- acting over a large area. Examples of the first case include the large class of protein-protein interactions that involve a do- main of a protein interacting tightly with a small peptide. The paradigm for this type of interaction is that of specific Src homology 2 (SH2) domains with specific small peptides con- taining a phosphotyrosyl residue. This interaction occurs with a dissociation constant as low as nM and is due to a specific binding pocket in SH2 domains not unlike a classical substrate- binding pocket (64, 205, 224, 225). Many other examples of domains that bind small peptides with affinities in the nano- molar to molar range have been described. The paradigm for the second case, i.e., surfaces that interact with each other over large areas, is that of the leucine zipper, in which a stretch of a-helix forms a surface that fits almost perfectly with another a-helix from another subunit protein (59, 161 also see refer- ence 4). Binding also occurs in the nanomolar range for such interactions (196). Other interactions may occur through in- termediate-sized complementary surfaces. It is evident that protein-protein interactions are much more widespread than once suspected, and the degree of regulation that they confer is large. To properly understand their signif- icance in the cell, one needs to identify the different interac- tions, understand the extent to which they take place in the cell, and determine the consequences of the interaction. This review is intended to supply an overview of three aspects of protein-protein interactions. First, we briefly describe a num- ber of physical, molecular biological, and genetic approaches that have been used to detect protein-protein interactions. Second, we describe several experimental approaches that have been used to evaluate the strength of protein-protein interactions. Third, we describe three well-characterized do- mains that are responsible for protein-protein interactions in a number of different proteins. As the literature on this topic is vast, we have not attempted to conduct an exhaustive review. Rather, we hope that this article serves as a journeyman���s guide to protein-protein interactions. The first and still the most comprehensive review on protein- protein interactions is that of Klotz et al. (116). This review contains a survey of the subunit composition and binding en- ergies of all oligomeric proteins that had been identified at the time, as well as a discussion of the geometry of interactions and an excellent discussion of the influence of binding constants, concentrations, and cooperativity parameters on the popula- tion of oligomers. A good discussion of channeling and com- partmentation is found in the monograph by Friedrich on quaternary structure (70) and the article by Srere (208). The review by Eisenstein and Schachman (57) contains an interest- ing discussion of the functional roles of subunits of oligomeric proteins and of approaches used to determine whether the monomers of oligomeric proteins are active. Also of interest is the discussion of proteins as machines (5) and a discussion of protein size and composition (78). PHYSICAL METHODS TO SELECT AND DETECT PROTEINS THAT BIND ANOTHER PROTEIN Protein Affinity Chromatography A protein can be covalently coupled to a matrix such as Sepharose under controlled conditions and used to select li- gand proteins that bind and are retained from an appropriate extract. Most proteins pass through such columns or are readily washed off under low-salt conditions proteins that are retained can then be eluted by high-salt solutions, cofactors, chaotropic solvents, or sodium dodecyl sulfate (SDS) (Fig. 1). If the extract is labeled in vivo before the experiment, there are two distinct advantages: labeled proteins can be detected with high sensitivity, and unlabeled polypeptides derived from the covalently bound protein can be ignored (these might be either proteolytic fragments of the covalently bound protein or sub- units of the protein which are not themselves covalently bound). This method was first used 20 years ago to detect phage and host proteins that interacted with different forms of E. coli RNA polymerase (177). Proteins that were retained by an RNA polymerase-agarose column (which was shown to be enzymatically active) but not by a control column coupled with bovine serum albumin were judged as interacting candidates. The interactions were substantiated in two ways. First, the 96 PHIZICKY AND FIELDS MICROBIOL. REV.

interaction of T7 0.3 protein with RNA polymerase was con- firmed by coimmunoprecipitation of the 0.3 protein with RNA polymerase antibody. Second, the interaction of T4 proteins with RNA polymerase was shown to depend on the form of RNA polymerase on the column: one T4 protein interacted with core RNA polymerase and T4-modified RNA polymerase but not with RNA polymerase holoenzyme, and another inter- acted only with the T4-modified polymerase. The phage pro- teins that bound RNA polymerase were identified by their absence in appropriate T4 and T7 mutants. Similar methods have been used, particularly by the labora- tories of J. Greenblatt and B. Alberts, to identify many other protein-protein interactions. Two excellent reviews on the topic, which cover many of the details of coupling and a num- ber of strategic considerations, have been published (69, 145). Candidate proteins can be coupled directly to commercially available preactivated resins as described by Formosa et al. (69). Alternatively, they can be tethered noncovalently through high-affinity binding interactions. Thus, Beeckmans and Ka- narek (14) demonstrated an interaction between fumarase and malate dehydrogenase by immobilizing the test enzyme with antibody bound to protein A-Sepharose, as well as by direct covalent coupling of the test enzyme to Sepharose. Some of the important considerations of a successful binding experiment are elaborated below. Purity of the coupled protein and use of protein fusions. An essential requirement for a successful protein affinity chroma- tography experiment is pure protein otherwise, any interacting protein that is detected might be binding to a contaminant in the preparation. Greenblatt and Li (80) did two experiments to establish that core RNA polymerase bound to NusA on the column rather than to a contaminant in the NusA preparation. First, they demonstrated that a fully active NusA variant pro- tein, which presumably contained different amounts of various contaminants (since it eluted at different positions in columns used to purify it), still bound core RNA polymerase second, they demonstrated by independent experiments that the com- plex contained equimolar amounts of NusA protein and core RNA polymerase. The easiest way to obtain pure protein, if the gene is avail- able, is through the use of protein fusions. Several such systems have been described in each case, the protein of interest (or a domain of the protein) is fused to a protein or a domain that can be rapidly purified on the appropriate affinity resin. The most common such fusion contains glutathione S-transferase (GST), which can be purified on glutathione-agarose columns (202). Other fusions in common use include Staphylococcus protein A, which can be purified on columns bearing immu- noglobulin G oligohistidine-containing peptides, which can be purified on columns bearing Ni21 the maltose-binding pro- tein, which can be purified on resins containing amylose and dihydrofolate reductase, which can be purified on methotrex- ate columns. (Other common protein fusions which add an epitope for the influenza virus hemagglutinin [12CA5] or c- Myc are also in common use and are used most often for coimmunoprecipitation [see the section on immunoprecipita- tion, below].) Purified fusion proteins are used in two ways to detect in- teractions on affinity columns. First, the protein is covalently coupled to the resins in the usual way, as was done by Mayer et al. (139) to detect a tyrosine-phosphorylated protein that bound to the SH2 domain of Abl tyrosine kinase and by Weng et al. (232) to demonstrate that the SH3 domain of c-Src binds paxillin. Second, the purified fusion proteins can be nonco- valently bound to the beads and then mixed with an appropri- ate extract or protein. This was done by Zhang et al. (248) to demonstrate an interaction of the N-terminal portion of c-Raf with Ras, by Flynn et al. (68) to detect the binding of an actin filament-associated protein to Src-SH3/SH2, and by Hu et al. (99) to demonstrate the binding of the SH2 domain of the p85 subunit of phosphatidylinositol 3-kinase to two different growth factor receptors. Influence of modification state. The interactions of many proteins with their target proteins often depends on the mod- ification state of one or both of the proteins (mostly by phos- phorylation). Thus, the recognition of Rb protein by the tran- scription factor E2F and by the transforming proteins simian virus 40 large T antigen, human papillomavirus-16 E7, and adenovirus E1A is more efficient with underphosphorylated than phosphorylated Rb (132, 133, 240). Conversely, SH2 do- mains of proteins, for example, recognize tyrosine phosphory- lated substrates several orders of magnitude more efficiently than they do their nonphosphorylated counterparts (64). Pro- tein-protein interactions that require a posttranslationally modified protein for interaction are not detected if the protein is purified by the use of expression vectors in cells in which the protein is not properly modified. A means to circumvent this problem is to use GST fusion vectors to express proteins in host cells more related to their origin. Thus, the interaction of bovine papillomavirus E5 oncoprotein with an a-adaptin-like molecule was confirmed by addition of beads to extracts of NIH 3T3 cells that were expressing the GST-E5 fusion (38). Similarly, a yeast GST vector that allows regulated expression of yeast GST fusion proteins has been described (148). Retention of native structure of the coupled protein. Failure to detect an interacting protein can result from inactivation of the protein during coupling. Ideally, coupling would immobi- lize a protein or a complex by randomly tethering it to the matrix through one covalent bond. For example, binding of E. coli proteins to immobilized l N protein occurred only when FIG. 1. Protein affinity chromatography. Extract proteins are passed over a column containing immobilized protein. Proteins that do not bind flow through the column, and ligand proteins that bind are retained. Strongly retained pro- teins have more contacts with the immobilized protein than do those that are weakly retained. VOL. 59, 1995 PROTEIN-PROTEIN INTERACTIONS 97

the cyanogen bromide (CNBr)-activated residues on the ma- trix were partially inactivated before coupling this was attrib- uted to the large number of lysine residues in l N protein and the generation of multiple (and denaturing) covalent bonds between l N protein and the matrix if the concentration of CNBr-activated matrix sites was too high (80). Therefore, de- termining that the coupled protein has retained its native struc- ture is an important control, when possible. With some pro- teins, such as RNA polymerase from E. coli, activity could be detected when the coupled protein was assayed on the matrix (177). With others, such as filamentous actin (F-actin) col- umns, the desired polymerized form was stabilized with phal- loidin (or by chemical cross-linking), and the proteins that bound F-actin were shown not to bind monomeric actin (14). Similarly, microtubule columns were stabilized with taxol (113). Native protein structure also depends on all subunits of a complex being present in the coupled resin. This can be as- sessed by SDS elution of a sample of the resin and comparison of the subunit composition of the eluted material with that of the starting material. In the case of E. coli RNA polymerase, all the components of the enzyme were still present (177). In the case of mammalian RNA polymerase II, one of the subunits did not reproducibly remain after coupling (206). Concentration of the coupled protein. To detect interactions efficiently, the concentration of protein covalently bound to the column has to be well above the Kd of the interaction. Thus, for the detection of weak protein-protein interactions, the concen- tration of bound protein should be as high as possible. Weak interactions can be completely missed on columns with lower concentrations of coupled protein, even if they contain corre- spondingly larger amounts of resin to maintain the same total amount of bound protein (see the sections on importance of characterization of the binding interaction and on binding to immobilized proteins, below, for a discussion of this point). Amount of extract applied. The amount of extract applied to the column can be critical for two opposing reasons. If too little extract is applied and the protein that binds is present at low concentration, too little protein will be retained to be detected, even if it binds with high affinity and is labeled with 35S (see, for example, reference 206). Conversely, if too much protein is applied, competition among potential ligands may result in failure to detect minor species. This was observed by Miller and Alberts (144) in looking for minor protein species that interact with F-actin. Other considerations. There are four distinct advantages of protein affinity chromatography as a technique for detecting protein-protein interactions. First, and most important, pro- tein affinity chromatography is incredibly sensitive. With ap- propriate use (high concentrations of immobilized test pro- tein), it can detect interactions with a binding constant as weak as 1025 M (69) (see the section on binding to immobilized protein, below). This limit is within range of the weakest in- teraction likely to be physiologically relevant, which we esti- mate to be in the range of 1023 M (see the section on limits of binding-constant considerations, below). Second, this tech- nique tests all proteins in an extract equally thus, extract proteins that are detected have successfully competed for the test protein with the rest of the population of proteins. Third, it is easy to examine both the domains of a protein and the critical residues within it that are responsible for a specific interaction, by preparing mutant derivatives (38, 216). Fourth, interactions that depend on a multisubunit tethered protein can be detected, unlike the case with protein blotting. One potential problem derives from the very sensitivity of the technique. Since it detects interactions that are so weak, independent criteria must be used to establish that the inter- action is physiologically relevant. Detection of a false-positive signal can arise for a number of other reasons. First, the pro- tein may bind the test protein because of charge interactions for this reason, it is desirable to use a control column with approximately the same ionic charges. Second, the proteins may interact through a second protein that interacts with the test protein although interesting in itself, the interaction may not be direct. Third, the proteins may interact with high spec- ificity even though they never encounter one another in the cell. The most famous example of this type is the high affinity of actin for DNase I (125). For all of these reasons, the prudent course is to indepen- dently demonstrate the interaction in vitro or, if possible, in vivo. Cosedimentation was used to confirm the interaction of RAP 72 (now known as RAP 74) and RAP 30 with RNA polymerase II (206), NusA protein with core RNA polymerase (80), and NusB protein with ribosomal protein S10 (138). In other cases, more biological criteria were used. For example, antibodies were generated against many of the proteins that interacted with F-actin (but not monomeric G-actin) on col- umns, and these were used to demonstrate that more than 90% of the corresponding proteins were localized with an actin-like distribution during mitosis of Drosophila embryos at the syn- cytial blastoderm stage of development (144). The identifica- tion of three yeast actin-binding proteins was confirmed in three separate ways: one of the proteins was shown to corre- spond to the yeast analog of myosin by virtue of a shared epitope another protein colocalized with actin cables and cor- tical actin patches, and overproduction of the third protein caused a reorganization of the actin cytoskeleton (53). In the identification of microtubule-associated proteins, two criteria were used to demonstrate the authenticity of the results (113). First, antibodies for 20 of the 24 candidate microtubule-asso- ciated proteins stain various parts of microtubule structures of Drosophila embryos during the cell cycle. Second, many (but not all) of the microtubule-associated proteins isolated on mi- crotubule affinity columns are the same as those isolated by traditional cosedimentation methods of Vallee and Collins (219). Failure to detect an interaction can occur for a number of technical reasons, described above. A false-negative result can arise for two additional reasons: the interacting protein may not be able to exchange with another protein to which it is binding, or the two proteins may not be able to interact both with each other and with the resin. Protein affinity chromatography does not always yield an- swers corresponding to other approaches. For reasons that are unclear, a large number of proteins were detected by probing SDS-polyacrylamide gel electrophoresis (PAGE) gels with a GST fusion of the SH2 domain of Abl tyrosine kinase, but only a couple of proteins were detected on columns coupled with this protein (139). Similarly, a specific protein was detected on F-actin columns stabilized by suberimidate cross-linking but not with phalloidin (144). Finally, G-actin interacting proteins are very difficult to detect with columns of G-actin, although such columns bind DNase I by contrast, DNase I columns can be used to detect such G-actin interactions (24). Affinity Blotting In a procedure analogous to the use of affinity columns, proteins can be fractionated by PAGE transferred to a nitro- cellulose membrane, and identified by their ability to bind a protein, peptide, or other ligand. This method is similar to immunoblotting (Western blotting), which uses an antibody as 98 PHIZICKY AND FIELDS MICROBIOL. REV.